MANAGED SERVICE PROVIDER“MSP”MARKET UPDATE Q3 2025NEW YORK|LONDON|LOS ANGELES|PARIS|MUNICHSAN FRANCI.
2025-11-03
18页
5星级
1 White Paper Potential Uses of AI in Federal Credit Prepared by Ugur Koyluoglu Prepared on Oct.29,2025 2 Table of Contents Table of Contents.2Table of Acronyms.3Abstract.51.Introduction.62.Private sector examples,case studies and fresh ideas on AI incorporation into federal credit programs.102.1.Application.182.2.Underwriting and due diligence.202.3.Loan management and monitoring.252.4.Broader program management.293.Generalized Recommendations.303.1.Make AI a priority.303.2.Develop an AI strategy.313.3.Set up key enablers.343.3.a.Governance.343.3.b.Technology and data.363.3.c.Monitoring,Testing,and Controls.373.3.d.Talent and Organization.383.4.Develop and implement a robust roadmap.383.5.Continue to iterate and monitor AI program.394.Appendix-AI Supremacy.424.1.Definitions and growing impact.424.1.a.Predictive AI.424.1.b.GenAI.434.1.c.Agentic AI.434.2.Rapid adoption and transformative power.444.3.Emerging trends and how AI is being used generally.484.4.Why now and implications of inaction.494.5.Financial services use cases.504.6.AI limitations and risks.534.7.Regulation.565.About the Author and Contributors.596.Get in Touch.59 3 Table of Acronyms Acronym Definition AI Artificial Intelligence API Application Programming Interface CCPA California Consumer Privacy Act FSS Factiva Sentiment Signals GDPR General Data Protection Regulation:the EUs data privacy and security law GenAI Generative Artificial Intelligence GPT Generative Pre-trained Transformer(e.g.,ChatGPT)HAI Human-Centered Artificial Intelligence(Stanford HAI AI Index reference)IP Internet Protocol LGD Loss Given Default LLM Large Language Model ML Machine Learning MRM Model Risk Management NLP Natural Language Processing PD Probability of Default RAG Retrieval-Augmented Generation RL Reinforcement Learning SOP Standard Operating Procedure AI SME Artificial Intelligence Subject Matter Expert 4 5 Abstract Artificial Intelligence(AI)is rapidly reshaping our world by offering substantial improvements in problem-solving and operational efficiency,which in turn create innovative working methods and enhance daily lives.Almost every company and organization whether public or private,for-profit or non-profit is making AI adoption a priority to drive efficiency,augment decision making,hyper-personalize offerings,improve internal management effectiveness,or reinvent business models.As stewards of taxpayer dollars,federal credit programs have an imperative to thoughtfully explore how AI can also enhance their processes and implement solutions making them more efficient,effective,and less error prone while mitigating potential risks and limitations associated with AI.This white paper is part of an ongoing project led by the Federal Credit Working Group to explore and discuss the potential uses of AI within federal credit programs.We anticipate a surge in public sector adoption of AI,given the rapid adoption of AI globally and the Trump administrations emphasis on enhancing efficiency,reducing fraud,waste and abuse,and shrinking headcounts.AI is already being used in various federal credit processes at select agencies,such as predictive modeling for risk assessment based on numerical data and chatbots for borrower support.However,there are significant opportunities to expand its presence in both ad-hoc use cases and structured applications across all agencies to transform the way federal credit professionals operate enabling them to boost operational efficiency,improve decision making,reduce credit losses,strengthen portfolio management,minimize fraud,and deliver superior services to borrowers.While the potential benefits are substantial,implementing AI in government lending programs also presents challenges.Among these are ensuring data quality,preventing bias in algorithms,attracting and retaining talent,upskilling existing staff,and designing,piloting and scaling solutions.There are also new risks to cybersecurity,and whether the public will trust AI.Based on private sector examples,potential uses of AI in federal credit include borrower relationship management,loan processing,co-pilot assistance for credit analysis and report generation,risk assessments using unstructured and large datasets,portfolio monitoring,fraud detection and prevention,compliance checks,translation of policy statements or processes to automation,research tasks,and overall information management,to name a few.The top five areas for AI introduction or expansion,based on the potential benefits,are:i)having an AI assistant or co-pilot for federal credit analysts for especially to conduct research,review documents and generate reports,ii)applying machine learning techniques to a combination of structured and unstructured data for better risk assessment and real-time portfolio monitoring,iii)reducing fraud by detecting anomalies and acting early,iv)improving the accuracy and efficiency of credit subsidy determinations,and v)increasing the use of chatbots to support loan applicants.This white paper presents industry observations,working examples,and fresh ideas,followed by a list of steps federal credit agencies can take to embark on their AI journey or accelerate their existing initiatives.Its aim is to help agencies and professionals effectively navigate the complexities of AI.6 1.Introduction Rise of AI AI powers chatbots for customer service,aids discovery of new drugs,personalizes shopping experiences,runs autonomous vehicles,and literally reimagines daily tasks for working professionals across the globe and has the potential to create one of the most significant forces for change in human history.Large businesses,startups,and governments alike recognize AIs significance and are investing hundreds of billions of dollars to quickly develop and adopt new AI tools and supporting infrastructures necessary to expand its implementation.In July 2025,the White House released a report titled Winning the Race:Americas AI Action Plan1 that touched upon how the nation could accelerate AI innovation,build AI infrastructure,and lead in applying AI to international diplomacy and global security.This initiative reinforces the strong commitment of the government to rapid technological advancement,while addressing security concerns and driving economic growth.The effort has momentum in the private sector as well:Almost every company and enterprise,both for-profit and non-profit,is making the adoption of AI a priority.At the same time,both private and public organizations need to understand AIs risks and limitations and incorporate the necessary controls and governance to ensure responsible AI adoption.AI for federal credit professionals The federal credit process is complex and time-consuming made up of many demanding components,including federal credit program awareness and outreach;application intake,screening,and evaluation;loan origination and underwriting;credit subsidy determination and budget analytics;disbursement;lender and guaranty management;loan securitization;portfolio management and reporting;property/collateral management;delinquent and default loan collections;account servicing and repayment,among others.Similar to the lending institutions in the private sector,which have been increasing their AI usage,AI could help federal credit professionals to improve efficiency,enhance decision making,and provide better services to borrowers.Federal credit professionals are already using AI today for some use cases such as chatbots for borrower assistance,research using large language models(LLMs),and predictive models applied to numerical data for risk measurement.However,we see several opportunities to expand AI applications in near-in use cases,inspired by successful private sector examples,including:Assistant or co-pilot for credit analysts:Reviewing documents,drafting reports,information gathering,speeding up and enhancing credit analysis,1 https:/www.whitehouse.gov/wp-content/uploads/2025/07/Americas-AI-Action-Plan.pdf,Executive Office of the President of the United States.(July 2025).Winning the AI Race:Americas AI Action Plan.Washington,D.C.:The White House.7 Risk assessment:Especially to extract risk signals in unstructured large data sets,Loan processing:Improving efficiency and accuracy in loan applications,Fraud,waste and abuse detection and prevention:By adopting AI-driven prevention and early detection systems that combine statistical anomaly detection,supervised learning,and rule-based controls,Compliance checks:Automating compliance,verification,and review processes,Automated data extraction,cleaning,integration and verification:Reducing processing time,and Code writing:Assisting in software development and automation.Predictive,generative,and agentic AI Predictive AI:These AI systems analyze historical data to make predictions about future events or outcomes.They use statistical techniques,machine-learning algorithms,and data mining to identify patterns and trends in data.While akin to some statistical work federal credit professionals are already doing for risk models,predictive AI extends the analysis using new algorithms.Generative AI:These AI systems,often called GenAI,create new content,based on the data/information input they are provided and prompts from human operators.They use algorithms to generate text,images,music,or other forms of media,aimed at resembling what humans might produce using the same inputs.They work with unstructured data exceptionally well.Among these,LLMs are highly relevant as these are designed to understand and generate human-like text,driven by extensive training on diverse text data.LLMs still require humans to craft the right prompt question to maximize the LLMs ability to generate accurate,relevant,and useful responses.Federal credit professionals will master prompt engineering over time as multiple prompt iterations with LLMs might initially be needed to get to targeted outcomes.As an extension of LLMs,deep learning is demonstrating an increasing capability to solve complex analytical problems,often achieving performance levels comparable to that of human experts.Agentic AI:These are streamlined AI systems and autonomous bots that are designed to operate relatively autonomously with limited human intervention.They can even make decisions without further human prompts and carry out complex activities independently.For example,an Agentic AI agent could use GenAI to produce customized reports,which include forecasts from predictive AI models and syntheses of information based on unstructured data all aligned with user objectives.While AI tools and agents are being adopted,human-machine collaboration is emerging as a superior approach to enhance productivity by combining human expertise,creativity and intuition with machine efficiency.Agentic AI and reinforcement learning(RL)play crucial roles in this synergy,allowing machines to learn from human feedback and adapt accordingly.This collaboration provides better explainability and ensures ethical standards are upheld in AI systems.8 Transformative power differentiated by the type of work The effectiveness of AI and its impact will differ based on the task,process,and type of work and level of sophistication and critical thinking required.Where AI is most effective tackling mundane tasks,documentation review,simple reasoning,and automatable workflows,we expect AI dominance to be strong and very fast,following a steep S curve over time similar to what has been seen with many AI-powered hybrid chatbots and virtual assistants that handle routine customer inquiries at some federal credit agencies.Where AI is currently less effective,for example,in tasks related to decisions that are higher stakes and require more critical thinking,we expect AI to still contribute some value and increasingly get better over time.Whats most important is for federal credit professionals to begin asking themselves how these tools can help them become better and faster at operational tasks and what becomes possible because of the existence of AI.Even for work and decisions that require more critical thinking,federal credit professionals need to explore how AI tools could be leveraged to improve outcomes.But AI is far from foolproof and federal credit professionals must continue to ensure these tools are used responsibly,within established safeguards and controls.Governance and AI risk management Alongside its transformative potential,AI models have inherent limitations,raise novel and complex risks,and can be prone to bias embedded in the training data used to generate AI algorithms.Ensuring data quality and accuracy is extremely important when developing and training AI models.LLMs have intrinsic limitations that inhibit them from being 100%reliable,and mistakes and so-called AI hallucinations are not fully avoidable.Although newer AI technologies are reducing potential issues and will continue to improve,intrinsic limitations and some weaknesses of AI models are insurmountable,at least in the current paradigm.All this reinforces the need for appropriate governance and management of AI risks,and introduction of necessary guardrails and controls.Organizations need to ensure that their AI tools operate safely and responsibly,within established laws,regulatory requirements,ethical boundaries,internal safeguards,and dynamic controls.Since federal credit is held to a high standard,it will be essential to establish strong governance and AI risk management processes.In the private sector,AI risk and governance are evolving rapidly.Leading firms are introducing both“top-down”governance frameworks and“bottom-up”risk assessments for specific AI use cases.These approaches could be adopted by federal credit professionals to ensure that AI models operate within appropriate safeguards that are consistent with an Agencys risk appetite.As described more fully in Section 3,top-down governance is built around five key building blocks:1.Determining the organizational AI strategy,including goals and priorities 2.Setting the rules of the road,or guardrails,for AI deployment 9 3.Establishing a risk framework for core risk concepts and roles and responsibilities across different functional groups 4.Developing risk assessment principles and criteria for evaluating individual use cases 5.Ensuring organizational reinforcement for education,training,resourcing,and other enablers Bottom-up risk assessment is derived from these top-down principles and is focused on reviewing,approving,controlling,and monitoring specific use cases.Both top-down and bottom-up frameworks need to be aligned with an organizations level of AI maturity.While simple approaches can be adopted for initial pilots,much more rigorous solutions will be required to support“governance at scale”for widespread operational rollout.Also,importantly,AI risk management at the early stage of AI adoption includes raising awareness for the implications of inaction.Federal credit programs must make sure the necessary infrastructure,strategy,people development and change management are in place,and their organizations are not falling behind on the latest thinking in the field.Lastly,reliance on technology should be balanced.There are already studies into the impact of cognitive offloading and the loss of skills associated with AI adoption.Programs should be aware that increasing delegation to AI can diminish independent thinking,problem-solving,and decision-making skills unless mitigations such as human-in-the-loop workflows,training,and validation requirements are put in place.Overview of the rest of this paper In this report,we present a comprehensive set of potential AI use cases for federal credit,supported by industry insights,illustrative case examples,and innovative concepts tailored to the sector(Section 2).We also translate those use cases into practical guidance and offer generalized recommendations and actionable steps federal credit agencies can take to embark on their AI journey or accelerate their existing initiatives(Section 3).Because agencies vary in mission focus,target audiences,product portfolios,technology and data maturity,organizational culture,talent and AI expertise,and available budgets,the guidance is intentionally generalized.Key recommendations include:Make AI adoption a priority and get going,if not already in progress Develop AI strategy and identify prioritized initiatives Set up key enablers such as governance,technology and data,monitoring,testing,and controls,talent and organization Develop and implement a robust roadmap Continue to iterate and monitor AI program If you are unfamiliar with AI,we encourage you to consult Section 4:Appendix.It contains essential background information,definitions,and context about predictive,generative,and agentic AI,as well as their use cases in financial services,associated risks,and regulation.This information will help you engage more effectively with the subsequent sections.10 2.Private sector examples,case studies and fresh ideas on AI incorporation into federal credit programs Given the widespread adoption of AI and vast improvement in accessibility through popular apps,including AI-based search engines,we are seeing an exponential increase in AI use cases.People are using GenAI for activities like content creation,learning and education,personal and professional support,technical assistance and troubleshooting,creativity and recreation,and research and analysis.Figure 1 lists 100 of the most common activities where individuals are benefiting from AI in their personal and professional lives.Figure 1:How people are using GenAI based on a review of multiple online sources 11 All of the mentioned uses in Figure 1,such as reviewing and editing documents,drafting reports,conducting specific research,or canvassing the web and find detailed information about the applicants company and project to understand their status and level of progress,are of course available for federal credit professionals and can be further tailored and optimized to the profession with the right set of prompts to accrue incremental benefits.By scaling and systematizing the use of AI throughout the organization,these ad-hoc applications can yield maximum benefits at the program level,particularly by replacing or streamlining routine,time-consuming,and rules-based tasks.Unlike individuals who tend to use off-the-shelf AI applications such as OpenAIs ChatGPT or Googles NotebookLM,corporations often build enterprise AI platforms to create rigorous secure environments that manage and restrict data flows.Also importantly,ability to integrate internal and external data sources,structured and unstructured information,is key to increase the quality and detail of output.Data privacy,reputational risk,and regulatory compliance are particularly relevant to federal credit lending programs,which require stringent adherence to applicant privacy.Any AI tool employed must implement necessary restrictions to safeguard proprietary information in accordance with the Trade Secrets Act,necessitating a secure,controlled,and trusted AI infrastructure.To get the reliable outcomes from AI even for ad hoc use cases role-based training,standardized prompt templates,and retrieval-augmented generation(RAG)anchored to internal sources will also be needed at a minimum.Role-based training should deliver tailored instructions basic safe-use and classification for all staff,prompt-engineering and verification skills for analysts,and secure integration/API training for developers so that each role can confidently use GenAI within its authority and risk profile.Prompt engineering has emerged as a crucial technique in optimizing the performance of AI models.The quality of a prompt significantly influences the effectiveness and relevance of the responses generated.High-quality prompts avoid ambiguity and are characterized by their completeness,clarity,and specificity.Completeness entails providing all necessary context and information to guide the AI effectively,while clarity is needed so that the intent and instructions are easily understandable.Specificity is also essential,as targeted prompts lead to more relevant and precise outputs,reducing the chances of misinterpretation by AI.Logical coherence in prompts enhances the overall interaction with AI,allowing for a more fluid and meaningful dialogue.The importance of prompt guidance cannot be overstated,as it equips users with the strategies necessary to construct prompts that yield high-quality responses to consistent needs,important for a wide range of uses of AI in federal credit.As federal credit professionals apply GenAI in their research,they should be aware that each use case requires thought,testing and iteration,and one excels in getting better outcomes from GenAI tools through experience.For example,retrieval and reasoning performance of AI models degrades significantly when the length of the documents increases.Through better prompts providing more context and grounding,refining AIs response iteratively,and 12 giving flexibility to AI about the length of the synopsis,one can get better results.Another approach is breaking the research task to manageable chunks.AI models have a maximum token limit,which constrains how much information can be processed at once.Once these limits are reached,AI model might focus on the beginning,middle,or end of the document,causing it to miss critical information.Also importantly,RAG augments a GenAI model with a retrieval layer that fetches relevant internal documents and passages,which the LLM then uses as evidence when generating answers.This ensures the models responses are grounded in verifiable agency documents and include citations,reducing hallucinations and improving trust for ad-hoc research.Federal credit case example 1 AI Toolbox The Department of Energy has developed the AI Toolbox(Joulix),a centralized accelerator designed to boost productivity and innovation.The Loan Program Office is currently piloting its capabilities and reviewing its processes to understand how to integrate Joulix to support and enhance federal credit operations.The three primary mission functions of the AI Toolbox are to:Accelerate:Advise,guide,and educate DoE offices on strategic AI applications and disseminate available tools Empower operations:Provide secure access to an integrated AI tool suite while maintaining and enhancing platform architecture.Design&deploy:Convert operational needs into user-ready and deployable AI solutions tailored to the departments specific requirements Joulix offers standardized guidance,robust technology infrastructure,and rigorous security measures for end-users to streamline credit-related processes,strengthen analysis capabilities,and empower operational efficiency.Joulix currently includes a curated catalogue of AI applications designed to meet diverse operational needs across federal credit workflows:13 These tools offer key benefits via substantial workflow efficiencies,analysis and decision-support capabilities for federal credit operations.Some examples include:Document review and analysis:Tools such as Fusion Docs AI,PDF Analyzer,and Volt Scribe AI enable rapid,detailed review of complex credit applications,feasibility studies,and compliance documents to review Report generation:Expert Grid AI and Cognitive Spark Engine facilitate the creation of in-depth analytical reports and strategic assessments,supporting robust credit risk evaluation and decision-making Operational productivity:EnerGPT automates routine content tasks and supports interactive query-driven workflows,freeing staff to focus on higher-value activities Joulix is currently powered by the advanced Gemini Flash 2.0 AI model and has been integrated with DoEs OneID system,providing secure authentication and consistent user experience across AI tools.Tool catalogue Description EnerGPT Basic ChatGPT-like app for online search,longform response,reasoning,improved prompts,and file uploads to enhance productivity by generating content,summarizing materials,and streamlining workflows PDF Analyzer Application to interrogate multiple PDFs with specific,targeted sets of questions to simultaneously analyze content Cognitive Spark Engine Application to facilitate ideation.Takes user inputs like topic,goals,questions,stakeholders and follow up clarifications as needed before providing a report-like analysis to simplify problem solving,decision making,and strategic planning with AI-driven outputs Expert Grid AI Produces expert-level reports by combining user input with AI-driven expert perspectives SmartPD Creator Creates comprehensive and accurate position descriptions for your organization Volt Scribe AI Chat application with a focus on discussing text(s)through an interactive interface to easily analyze text to uncover deeper meaning,ask context-aware questions,and refine text in an interactive environment Fusion Docs AI Similar to PDF Analyzer,but single-shot questions and generally longer-form responses that generate longer and more reasoned outputs PWS Builder Generates clear and compliant performance work statements In what follows,we focus on more involved professional uses and macro-applications in credit underwriting and risk management that will substantially improve efficiency,effectiveness,and accuracy in federal credit.To illustrate the potential uses and the opportunities to rethink and re-design credit processes(targeting agentic solutions that pull the data and information,verify parts of all of it with comparisons to various datasets,review checklists for eligibility and suitability,run predictive models,analyze applications to then generate insights and reports),this chapter presents tangible examples of AI applications and services from the private sector along with case studies and new ideas in the federal credit context.We structured the discussion in four broad categories of loan application,underwriting and due diligence,loan management and monitoring,and broader program management processes,as shown in Figure 2.14 Figure 2:High-level lifecycle in federal credit Within these four broad categories,we include commonly referred to aspects of federal credit program management including program awareness and outreach;application intake,screening and evaluation;loan origination and underwriting;budget analytics(FCRA,Credit Subsidy,etc.);disbursement;lender and guaranty management;loan securitization;portfolio management and reporting;property/collateral management;delinquent and default loan collections;and account servicing and repayment,as described in Table 1.Table 1:Potential Uses of AI across federal credit lifecycle Lifecycle Activities High level description Potential Uses of AI(ideas)1.1 Application 1.Program Awareness and Outreach Advertise the program Identify and reach out to potential borrowers Answer potential borrower questions Generate informational content about the program,refine,and streamline this content and all relevant marketing material Conduct data analysis to identify potential borrowers that would be a good fit for program Automate reaching out to potential borrowers and underserved populations 2.Application Intake,Screening and Evaluation Communicate better with potential borrowers Accept and log applications Screen for completeness of applications Evaluate for eligibility and suitability before detailed review Pre-qualification and guidance Automated data extraction,integration,cleaning,and verification Anomaly detection especially to identify fraud Improve the ability of borrowers to shop for federal loans and to determine the qualifications they must meet to obtain a federal loan Better and cheaper borrower engagement using AI chatbots on the landing page,possibly improved customer experience Organize and summarize application materials Auto-flag incomplete or ineligible applications Make auto-recommendation on eligibility Increase fraud detection Reduce manual entry/errors,faster processing times 2.3Loan Managementand Monitoring2.2Underwriting and Due Diligence2.1Application2.4 Broader Program Management 15 Lifecycle Activities High level description Potential Uses of AI(ideas)Translate policy statements,processes and compliance rules into code 1.2 Underwriting and Due Diligence 1.Property/Collateral Management Gather information on the collateral asset(s)Assess existence and legal ownership of the collateral Confirm the value of pledged assets(real estate,equipment)to mitigate credit risk(and reassess as needed)Collect information from unstructured data and large data sets,analyze data for pledged assets automated collateral valuations Input satellite imagery with AI to monitor condition or existence of collateral assets(such as.real estate)2.Lender and Guaranty Management Define,screen,and review for lender eligibility Write and align on lender agreements including terms of lender participation agreements,performance standards,reporting requirements,and terms applicable to loan services.Review lender and servicer performance Draft lender agreements Apply advanced scoring algorithms to evaluate lender performance and detect noncompliance patterns,if any graphs/networks to understand inter-dependencies/potential contagion 3.Loan Origination and Underwriting Determine the creditworthiness of borrowers Establish loan terms that balance demand,program objectives with risk exposure (If needed)Conduct due diligence to validate underwriting assumptions Generate and sign loan agreement document Confirm covenants met prior to disbursement Enhance credit worthiness assessment and improve credit scoring/rating models with a more comprehensive set of risk indicators Apply advanced algorithms for risk quantification from unstructured large datasets such as news and social media Translate policy statements,processes and compliance rules into code Fast track compliance assessments Generate loan agreement document draft Review materials to confirm covenants met and make recommendation if met pre-disbursement Increase efficiency by automating data mining,fact synthesis,documentation,and credit underwriting content creation 4.Loan securitization Assess loan risk and pool assets Analyze loan pool risk and optimize how assets are grouped 16 Lifecycle Activities High level description Potential Uses of AI(ideas)1.3 Loan Management and Monitoring 1.Account Servicing and Repayment Generate and send repayment reminders and log repayments Communicate with and answer questions from recipients Generate draft reminders/notices Automate distribution Proactively identify borrowers at risk of delinquency for proactive engagement 2.Delinquent and Default Loan Collections Generate and send delinquency reminders/notices Contact borrowers to collect on delinquent loans Generate draft reminders/notices Collection/workout co-pilots 3.Portfolio Monitoring and Reporting Continuously monitor portfolio performance and risks Early detect financial distress Pattern recognition and anomaly detection Generate reports on portfolio Improve borrower engagement via automated communications Continuous better,faster,and cheaper monitoring of credit signals from unstructured data from such sources as news,social media,court filings Enable proactive risk management and potentially reducing losses Apply anomaly detection models to flag unusual repayment trends and fraud indicators Hyper-personalized communications 4.Budget Analytics Calculate credit subsidy Conduct credit subsidy re-estimates Use machine learning to improve accuracy of input parameters to credit subsidy calculation,including improved prediction of likelihood of default Increased quality of agency forecasts of creditworthiness of potential borrowers 1.4 Broader Program Management 1.Program Policy and Evaluation Define and document program objectives and policies Identify potential indicators to track progress toward objectives Conduct periodic,results-oriented evaluation on program outcomes and effectiveness Develop program evaluation design Use natural language processing(NLP)models that generate coherent and contextually relevant text based on prompts provided by users to analyze program feedback,audit reports,and regulatory changes Draft program evaluation report 2.Daily operational tasks Various other activities Draft emails and memos Conduct desk research Summarize documents and meeting notes Write code 17 Across all elements of the federal credit lifecycle,AI-powered tools can revolutionize policy and process management by translating complex,natural language statements into executable automation code,a practice often called Policy as Code or Rules as Code.This capability significantly accelerates the conversion of written business rules,regulatory requirements,and operational procedures for any process into automated workflows,eliminating much of the manual work and coding effort.It also minimizes human error and establishes consistent enforcement of policies across an organization.Each of these AI use cases requires significant data and information.For example,assessing the creditworthiness of a corporate borrower through AI modeling necessitates a diverse array of data,not only about the borrower but also historical data about comparable borrowers and their performance.Financial data forms the backbone of this analysis,including detailed balance sheets,income statements,and cash flow statements,which provide insights into the borrowers financial health,profitability,debt-service coverage,and liquidity.Historical data,credit scores from leading agencies,and operational metrics like revenue streams are essential for understanding a companys past performance and that borrowers ability to manage debt.Complementing this quantitative data,qualitative insights into management quality and corporate governance are crucial for assessing the overall risk profile.In addition to traditional financial metrics,a comprehensive creditworthiness evaluation incorporates industry-specific benchmarks,macroeconomic indicators,and alternative data sources.Monitoring market trends,economic conditions,and competitive positioning offers insight into external factors that could impact the borrowers sustainability.Alternative data,such as sentiment analysis from news articles and social media,can provide an emerging perspective on most recent developments,public perception and potential risks.By employing advanced AI techniques such as machine learning and natural language processing,federal credit analysts can synthesize these multifaceted data points,enabling a more nuanced understanding and quantification of a borrowers credit worthiness.The implementation of these AI use cases requires careful planning and resourcing.Each initiative varies in timeline,ranging from a few weeks to several months,and necessitates different internal and external resources.For instance,deploying well-established vendor tools such as customer support chatbots may take only weeks for design,development,and integration.In contrast,predictive analytics for assessing borrowers creditworthiness involve complex data management and require a diverse team,including data scientists,underwriting experts,and financial analysts,typically unfolding over medium to long-term timelines.Additionally,the models must undergo independent model validation.Federal credit agencies have significant opportunities to enhance the accuracy of their models and operations through data/information sharing across agencies as well as standardization and potential shared services and shared models.For example,there is a compelling need for investment in Automated Standard Operating Procedure(SOP)Analysis and Mapping.Currently,each agency such as the Small Business Administration,US Department of Agriculture,Department of Housing and Urban Development,or Veteran 18 Affairs maintains separate standard operating procedures(SOPs),handbooks,and notices that are often lengthy and subject to frequent updates.This fragmentation presents challenges for lenders and users alike.By leveraging AI,federal credit agencies can transform their disparate SOPs into a cohesive“Federal Credit Rulebook.”Such an initiative would streamline compliance processes,eliminate duplication,minimize friction in lending,and provide greater alignment among various federal programs,all while respecting the distinct missions of each agency.This approach not only enhances efficiency but also creates real economies of scale.Furthermore,private lender SOPs and overlays can be incorporated into the new rulebook to achieve economies of scale across the entire ecosystem.2.1.Application In this stage,AI can help potential borrowers to understand loan programs better and submit their applications along with the necessary documentation.It can also help federal credit marketing teams to develop analytical strategies for targeted outreach,and federal credit professionals to review the completeness and accuracy of the information gathered from the applications and assess the eligibility and suitability of the applicants.For applicants,AI chatbots improve the application process by answering questions 24/7 and proactively providing information and resources to help fill out applications.Federal credit case example 2:Chatbots Many federal credit agencies already deploy chatbots to improve customer experience and simplify access to loan and grant programs.Current implementations range from deterministic,rule-or tree-based systems to hybrid solutions that incorporate intent classification and retrieval components by AI.These deterministic/hybrid systems represent a strong starting point:they deliver predictable,auditable behavior and are easier to validate for compliance.GenAI chatbots which produce more flexible,conversational responses and advanced summarization are emerging and will likely be adopted progressively.Agencies should therefore pursue a phased approach that leverages existing deterministic capabilities while preparing governance,technical controls,and monitoring for responsible GenAI integration.Representative examples of chatbots used in federal credit include:Small Business Administration“SBA Chatbot”:The SBA implemented a chatbot that assists small business owners with inquiries about various loan programs,including the Paycheck Protection Program and Economic Injury Disaster Loans.The chatbot provides information on eligibility,application processes,and funding options,helping borrowers navigate the complexities of federal loan programs.19 Federal Student Aid “FSA Chatbot”:The FSA implemented a chatbot to help students and borrowers with questions related to federal student loans and financial aid.The chatbot provides information on loan programs,application processes,repayment options,and eligibility criteria,assisting borrowers in navigating their financial aid options.Department of Agriculture “USDA Rural Development Chatbot”:The USDA introduced a chatbot to assist individuals interested in USDA loan programs,such as the Single-Family Housing Guaranteed Loan Program.The chatbot can answer questions about eligibility,application processes,and available resources for rural development loans.There are several companies that provide AI chatbot implementation services and federal credit programs could explore the options,if they dont currently use chatbots.When selecting a vendor,federal credit agencies should consider factors such as integration capabilities,security features,ease of use,and quality of interaction with borrowers,among others.Synthesizing the market of potential borrowers and determining where a programs offerings are more attractive and where applicant pool is under-represented based on program design parameters is labor intensive and evolves over time.AI could help prioritize the use of limited resources in targeting outreach efforts.For instance,predictive AI tools can help agencies pinpoint individuals or businesses who are most likely to benefit from federal credit programs and with the highest propensity to apply for various offerings by analyzing large datasets and engagement metrics.An example of this in the private sector is Rocket Mortgages subsidiary Lendesk,which has introduced an AI assistant for mortgage professionals that streamlines the process of matching clients with personalized mortgage recommendations.2 A similar approach could allow federal credit organizations to be more strategic in their investments in programs.For example,in the context of USDAs Single Family Housing Direct Loan Program,AI models could analyze geographic data,income brackets,family size,and rural housing trends to flag households most likely to be eligible that have not yet applied.3 This potentially enables targeted outreach to underserved communities and amplifying the impact of the federal credit program.We also expect to see substantial benefits to implementing GenAI tools in initial scans of submissions to gauge eligibility and suitability of applicants.Once an application has been submitted,federal credit professionals review all documentation provided by applicants to determine if the applicant is eligible and/or suitable for the program against a set of criteria and evaluate whether the application is complete.GenAI tools can help speed-up and 2 Rocket Mortgages AI Technology:The Future of Mortgage Lending|Forbes 3 Single Family Housing Direct Home Loans|Rural Development 20 automate document organization,conduct eligibility and/or suitability checks,and detect fraud.GenAI can read and understand submitted documents,extract key information from these documents such as income statements,tax returns,and identification papers,making it easy for federal credit professionals to access the data they need without the hassle of manual sorting,4 categorize them by document type and help federal credit professionals complete a checklist to confirm the application is complete.This saves time and prevents anything from slipping through the cracks during the eligibility and/or suitability review process.GenAI can generate a report in a specific format and length with a recommendation on whether the applicant meets eligibility and/or suitability criteria.GenAI can support recommendations with citations in the documents and other data sources,so that federal credit professionals can quickly review and confirm eligibility assessments.There are several vendor tools to automate the extraction,verification and analysis of financial documents such as bank statements and tax returns for lenders.These platforms quickly process and organize the information,reducing the need for manual data entry and minimizing errors when determining if applications are complete,eligible and suitable.Finally,scanning data for eligibility and/or suitability of applicants is an opportunity for federal credit programs to identify and stop potentially fraudulent applications.Traditional methods of fraud detection can sometimes miss red flags,leading to the approval of applications that should have raised concerns.Predictive AI algorithms,on the other hand,analyze patterns in large datasets more effectively and detect anomalies in application patterns,submission timing,and inconsistencies across supporting documents that might indicate fraudulent activity.This can help government agencies proactively detect and prevent fraud,safeguarding taxpayer dollars.For example,AI models flag duplicate applications submitted under different identities or detect unusual patterns in income reporting that deviate from typical borrower profiles.These models also review alternative content in social media and news to identify warning signals,which help credit agencies intervene earlier in the process.2.2.Underwriting and due diligence Assessing credit worthiness During the underwriting and due diligence stages,credit analysts conduct a comprehensive evaluation of the borrowers creditworthiness and the specifics of the loan or guarantee.This process includes verifying information,analyzing financial statements and assessing risk factors such as debt service coverage ratio.Credit risk of the obligor and quality of the collateral are assessed and scored using templates and models.Traditional credit scoring methods often rely on statistical methodologies and are driven by a limited set of quantitative and qualitative risk factors e.g.a total of 10 separate factors which can be improved.4 When to Use AI-Powered Document Data Extraction|Foxit 21 Credit scoring models based on machine learning methodologies such as gradient boosting,random forests,and neural networks incorporate a larger number of risk factors into modeling and capture previously uncaptured interactions across factors and other non-linear phenomena.These models also optimize segmentation of models automatically from the underlying data.Moreover,these models can analyze vast amounts of data from additional and alternative content such as payment histories,transactional data,court filings,news,and social media activity.In the federal credit context,such credit scorecards and probability of default(PD),loss given default(LGD),exposure at default,and prepayment tools are key inputs to risk assessment,loss forecasting,current and expected credit losses,stress testing and credit subsidy calculation.The acceptance of credit scoring in the 1980s and 1990s provides a useful roadmap for evaluating the adoption of AI in federal credit programs administered by the SBA,USDA,HUD FHA,VA,and others.Credit scoring helped bring efficiency and consistency but also raised issues of fairness,transparency,and regulatory adaptation.These lessons are directly applicable to the current push to adopt advance machine learning and leverage GenAI for decision making in federal credit agencies.Federal Credit case example 3 Credit models Credit scorecards and regression-based models built on structured inputssuch as FICO/VantageScore,loan-to-value,borrower credit history,income,balances and payments,and macroeconomic indicators for retail lending;and financial statements,credit ratios(including debt-service coverage),industry outlook,and qualitative management factors for corporate lending and project financeare foundational to underwriting and portfolio monitoring in federal credit.These models are calibrated to a PD to estimate the likelihood a borrower will fail to meet obligations within a defined horizon.Federal housing programs(FHA,VA,USDA)and federally supervised mortgage markets rely on PD models alongside predictive models for prepayment and LGD as essential inputs to loan-level cash-flow projections,underwriting,reserving,capital allocation,and risk-based pricing.FHA,VA,USDA actuarial and program guidance and supervisory disclosures(including those related to FHFA,GSEs)describe elements of these modeling practices.There are other federal loans programs(e.g.Department of Education)that have used regression models,but the disclosure requirements vary.One input to many federal credit models is the FICO score,which reflects a weighted combination of credit-behavior componentspayment history(35%),amounts owed(30%),length of credit history(15%),new credit(10%),and credit mix(10%).FICO historically used traditional statistical techniques;in recent years FICO and other scoring vendors have reported incorporating ML techniques for specific products and enhancements.The precise extent and placement of ML across different score versions is not fully public.22 For federal credit,transitioning from heuristic scorecards and regression frameworks to ML is a logical,lower-friction step than moving to unstructured-data systems.ML methods(for example,gradient boosting,random forests,and neural networks)can use the same structured inputs to capture nonlinearities and interactions,potentially improving predictive accuracy and surfacing early-warning signals.That said,ML adoption requires rigorous feature engineering,validation,deployment,monitoring,governance,and documentation,and supervisors and end-users expect robust explainability and model controls.GenAI that ingests unstructured datadocuments,images,free text,or audiocan enable richer borrower insights(for example,automated document validation or anomaly detection in transcripts)but entails materially greater technical,operational,privacy,and compliance challenges.Federal credit agencies should prioritize adopting predictive ML for near-term,defensible gains using existing structured data and governance frameworks,and plan to introduce GenAI more gradually only after implementing strong controls for fairness,privacy,explainability,and operational resiliency.Numerous banks,private lenders and rating agencies have developed predictive AI-driven tools that help these institutions measure risks more accurately,with increased explanatory power and move through the credit review process more efficiently.For example,Zest AI is an AI solution that allows financial firms to utilize AI in credit underwriting.5 Zest AI uses machine learning to assess applicant data such as income,credit score,employment history,market trends,and property evaluations more quickly and accurately than manual methods.Zest AI has partnered with companies like Equifax,Experian,Fidelity National Information Services,and Fiserv to provide these services.6 Credit Research Assistant/co-pilot Similar to how AI can be used in the application review stage to quickly extract data from documents,and analyze and synthesize information,the technology can also be used in the underwriting phase.By applying NLP,LLMs,and ML algorithms onto borrower information such as annual reports,deal team assessments,and transcripts of applicant calls,several banks,private lenders,and vendors have already developed AI co-pilots and AI agents to support their credit workflows,analysis and report generation.7 In a typical commercial credit review process,a credit analyst,relationship manager,and risk analyst will need to digest more than 600 pages of information and generate more than 40 pages of output and analysis.Historically,this has been a manual process and could take teams multiple days and even weeks depending on the company size,loan size,and amount 5 Solutions|Zest AI 6 Partners|Zest AI 7 AI in loan underwriting:Use cases,technologies,solution and implementation|LeewayHertz 23 of information to process.GenAI credit co-pilots help to streamline this process and automate the credit review process across each of the key steps in the credit review process.These tools are especially helpful in drafting the credit memo,which is typically up to 20/30 pages of information that includes the investment thesis,supply chain risks,competitor landscape,financials,and more,augmenting analyst work and explaining where in documents it is finding the information so an analyst can quickly review and confirm its synthesis.Such tools can save analysts up to 60%of their time drafting the credit memo for a new client.Additionally,such tools can provide a confidence score based on the certainty the model has with its synthesis of the underlying information,so the analyst knows what areas require more human review as seen in Figure 3.Figure 3:AI-enabled credit report generation Leading banks,private lenders,rating agencies,and vendors have already developed and implemented AI agents to support their credit analysis and report generation.For example,Moodys Research Assistant offers a public example of AI-facilitated risk due diligence for organizations.This assistant aids in data collection,including credit ratings,financial statements,market data,macroeconomic indicators,industry research,default and recovery data,regulatory information,and geopolitical data.Integration of GenAI into an AI assistant further enhances the diligence process as AI will then parse through the amassed information to develop a comprehensive risk overview.Key functionalities include providing detailed financial analysis based on borrower performance,credit rating insights,and peer comparisons.Following the data collection and preliminary insights,the AI research assistant finishes the diligence process by creating credit risk memos,as illustrated in Figure 4,which synthesize all relevant data sources.The exact time savings from using Moodys Research Assistant,for example,can vary widely depending on the specific tasks,the prior processes used,and the complexity of the analysis being conducted.However,studies and user feedback generally suggest that users can save between 50%and 70%of the time typically spent on manual data gathering by using automated data collection features.24 When it comes to the time required for analysis and report generation,users can typically save between 30%and 50%.Overall,users may experience total project time savings ranging from 20%to over 60%,depending on the specific context and usage patterns.8 Figure 4:Moodys Research Assistant Another rating agency example is S&Ps CreditCompanion which is an AI tool designed to enhance the accessibility and usability of S&P Global Ratings data and research based on GenAI,NLP,LLMs and a retrieval augmented generator to facilitate its functions.CreditCompanion features a chat interface and streamlines access to data,allowing users to quickly find relevant insights and reducing the time spent on information retrieval.By supporting both structured and unstructured queries,the tool enables users to uncover data trends and qualitative insights in one centralized platform.Internal testing by S&P analysts showed that CreditCompanion can reduce the time spent on research by up to 40%.9 In addition to its efficiency,CreditCompanion offers reliable and transparent information by linking users directly to relevant source documents from S&P Global Ratings.This feature fosters trust in the insights provided,which is crucial for making sound financial decisions with conviction.Fitch Ratings also utilizes predictive and GenAI tools to improve efficiency and productivity,especially in software maintenance and development.Specifically,they are leveraging Amazon Q Developer and Amazon Q Transform for code generation and infrastructure modernization,leading to a reported 40ficiency gain and 20%improvement in developer productivity,respectively.10 Another public example,Auquan,is a venture-backed startup focused on providing AI solutions to automate labor intensive workflows for financial services firms.Auquans solution transforms the traditional process of manually reviewing financial documents into a 8 Moodys Launches Moodys Research Assistant,a GenAI Tool to Power Analytic Insights|Moodys 9 S&P Global Introduces CreditCompanion,Enhancing RatingsDirect on S&P Capital IQ Pro with Advanced AI Technology|S&P 10 Fitch Ratings gains 40ficiency with AWS gen AI tool|Bank Automation News 25 streamlined operation that generates comprehensive credit reports in minutes.11 By processing vast amounts of financial data quickly and offering analyses,the tool can dramatically cut report creation time from days to minutes,increasing team productivity and allowing analysts to focus on complex risk decisions.Such NLP tools can also be used in the loan disbursement phase to confirm whether key covenants are met prior to receiving funding.AI is not only helpful in commercial risk diligence.It can also help establish personal fact bases that generate insights for credit professionals.By aggregating and analyzing data from multiple sources,such as tax returns,payroll data,bank statements,and credit histories,AI systems provide a comprehensive view of a borrowers financial health and behavior.What differentiates this from traditional data aggregation is AIs ability to analyze these disparate datasets at scale,detect non-obvious correlations and dependencies(for instance,between part-time employment and likelihood of default),and generate predictive risk profiles in real time.Replicating this capability is not possible using manual processes or simple dashboards.For example,in the context of student loan programs,AI could combine Free Application for Federal Student Aid(FAFSA)data,academic records,part-time employment income,and spending patterns to assess a borrowers likely repayment ability post-graduation.This provides federal credit professionals with a more dynamic and forward-looking understanding of borrower risk,beyond static income or credit history.AI can also be used to help with property/collateral management as well as lender and guaranty management.For instance,AI can process satellite imagery and property records to assess the existence and condition of collateral assets.This enables early detection of asset risk without requiring in-person inspection.Furthermore,AI-powered risk models can evaluate lender performance and flag patterns of noncompliance or elevated credit risk.2.3.Loan management and monitoring AI can make a big impact in monitoring the creditworthiness of borrowers in real-time,performance of loans,risk management and compliance assessments,identifying waste,fraud,and abuse,and servicing the loan.Real-time identification of credit warning signals from unstructured data Using AI,federal credit organizations can identify early warning credit signals from unstructured data and improve the risk monitoring process.Traditional monitoring methods often rely on structured data that is updated from time to time,such as quarterly and annual reports and updated credit scores.This method may not capture the full picture of changing dynamics of a borrowers health.AI systems,on the other hand,can also analyze unstructured and almost continuously updated data sources,including social media activity,news articles,customer feedback,and supply-chain issues to identify potential risks and emerging trends.For instance,NLP 11 AI Agents for Institutional Finance|Auquan 26 algorithms scan news feeds and social media platforms for mentions of borrowers or industries,understanding the context and detecting sentiment shifts that may indicate a change in creditworthiness in real-time.Dow Jones Factiva Sentiment Signals(FSS)tool,powered by Oliver Wyman,is another example of how AI can enhance loan and portfolio monitoring through advanced analytics and deep learning,where a deep learning algorithm provides proactive signals of corporate risk events,anticipating potential downgrades or defaults three to six months ahead of time.Objectively analyzing and rating news information from over 10,000 media sources every 15 minutes,FSS calculates daily scores for more than 450,000 public and private entities globally,to detect sentiment shifts and early warning signals for proactive risk mitigation as illustrated in Figure 5.12 Figure 5:How the FSS tool works 12 Factiva Sentiment Signals API|Dow Jones Developer Platform 27 Figure 6 illustrates how FSS utilizes GenAI to analyze news articles and create real-time predictive signals and output for a select company 23andMe Holding Company.Elevated scores and the increases in the scores are credit warning signals which are further explained by attribution to specific news and articles for the credit analyst to review below the time-series chart.The time series chart(Figure 6)shows in blue 23andMe holdings run to downgrades and default(higher FSS score indicates higher likelihood of a downgrade or default)along with the count of good news and bad news.Figure 6:FSS and 23andMe use case example 100Positive articlesHighVery highModerateLowVery low806040202020202120222023December 2023FSS risk score 30(high risk).January 2024FSS risk score crosses 40(very high risk).February 202423andMes share price keeps on falling in the wake of hacks and losses.20242025010008006004002000200400ABCDEACEBDMarch 2025Bankruptcy filing.F6 months24 monthsNegative articlesFSS scoreLow articlesF 28 Loan performance,compliance and fraud detection AI can also help fast track ongoing compliance assessments.Since federal credit program borrowers must meet the covenants and terms set out in their loan agreements,federal credit professionals must monitor borrower compliance,which is often a time-consuming and resource-intensive task.GenAI can speed up this process by automating the analysis of relevant data,which includes publicly available information as well as data submitted regularly by lenders as part of loan agreements,and flagging potential breaches in real-time.For instance,where loan agreements are bespoke and complicated,AI can read through the loan program agreement and identify all the borrowers compliance obligations and create a compliance checklist for federal credit officers to use.AI can also scan information when borrowers submit regularly as part of their loan agreement and pre-populate compliance checklists with recommendations or flags the possibility of noncompliance is detected.Federal credit agencies can reduce fraud,waste,and abuse by adopting AI-driven prevention and detection systems that combine statistical anomaly detection,supervised learning,and rule-based controls.13 14 Anomaly detection helps federal credit agencies find suspicious activity potential fraud,waste,or abuse by flagging cases that deviate from normal patterns rather than relying only on fixed rules.Models trained on historical application,payment,and investigation data can uncover subtle signals(synthetic identities,repeated submissions from the same device/IP,or odd attribute combinations).This kind of network analysis can reveal coordinated fraud rings by linking accounts,phones,addresses,and transactions.Detected anomalies are prioritized for investigator review,reducing false positives and focusing resources on the highest-risk cases;logged findings enable continuous improvement and maintain auditability and compliance.In lending and investment businesses,prevention at intake and pre-disbursement is essential to stop improper payments before they occur.Real-time risk scoring that combines identity verification,device and geolocation signals,document forensics,and anomaly scores can block or flag high-risk applications for secondary checks.NLP can detect inconsistent free-text statements or known red-flag phrasing,while automated duplicate and pattern detection reduces waste by catching erroneous or redundant payments.Integrating these AI checks into workflows holding disbursements,requiring additional verification,or routing cases to investigators materially reduces fraud losses and curbs wasteful or abusive claims with minimal friction for legitimate applicants.AI-powered fraud,waste,and abuse programs must be built on rigorous governance,with clear rationale and human-in-the-loop oversight,for them to be effective and legally defensible.Agencies should maintain auditable logs of model decisions and continuously monitor precision versus false-positive tradeoffs so they can tune performance responsibly.13 AI seen as critical tool to fight fraud,waste and abuse,even as it empowers scammers|Thomson Reuters Institute 14 https:/ Investigator outcomes would need to be incorporated into model updates to improve accuracy and reduce unnecessary investigations over time.Privacy protections,bias testing,and strict access controls are essential to ensure enforcement is fair and compliant with federal records and privacy obligations.Agencies could begin by strengthening existing rule-based controls with AI-powered insights for immediate impact and then expand to real-time scoring and advanced network detection as their data integration and oversight capabilities mature.AI can also help review the paperwork that federal credit borrowers often need to submit on an ongoing basis.Utilizing NLP and ML,GenAI can automate the extraction and analysis of key information from various documents to significantly reduce the time and effort required for manual reviews,enabling federal credit professionals to focus on more complex evaluations that necessitate human judgment.2.4.Broader program management Outside of the investment process,federal credit programs also conduct broader program management activities.These include setting the program policy,plans,milestones,resources,and evaluating progress against these as well as day-to-day operational tasks.A number of memos,reports and dashboards need to be produced.AI can make these broader program planning and management activities more efficient and effective.In terms of program policy and evaluation,GenAI can help develop program evaluation design,NLP models can help analyze program feedback and data,and GenAI can help draft evaluation reports.In terms of daily operations,AI can help federal credit professionals track and update workplans,summarize meeting notes,draft emails,prepare reports,and research answers from internal documents.15 AI can also help professionals research answers to questions from internal documents.In the commercial space,IBMs Watson Assistant is used by companies to improve customer service and internal operations.Watson Assistant can be integrated into various business processes to provide employees with instant access to information,automate routine inquiries,and assist in decision-making by analyzing data patterns.16 For instance,Watson Assistant is used in call centers to guide agents through customer interactions,ensuring they have all the necessary information to resolve issues efficiently.In the government agency setting,an example of a virtual assistant in use is Curie,the General Services Administrations ServiceNow Virtual Agent.17 Curie uses machine learning to provide predictive results for chat entries.It utilizes knowledge-based articles to improve IT service requests from employees,allowing simpler queries to be addressed faster and freeing up more human time for complex service requests.15 Real-world gen AI use cases from the worlds leading organizations|Google Cloud Blog 16 IBM Watson|IBM 17 Artificial intelligence|GSA 30 3.Generalized Recommendations Based on private-sector observations,federal credit case examples,and fresh ideas explored in this paper,predictive,generative,and agentic AI offer innovative solutions available to all aspects of the federal credit life cycle,from application processing to portfolio monitoring.While all federal credit agencies facilitate access to credit,their specific focuses and target audiences set them apart.Since their size,product offerings,data and technology environments,culture,talent,and expertise in AI,and budgets for development are all different,we can only provide generalized recommendations based on learnings from industry observations in this paper.Federal credit organizations regardless of their level of maturity in AI adoption,tech and data environment,and talent and expertise should(re)commit to AI adoption and implement a strategic and well-funded approach that aligns with their goals and resources.To effectively start or accelerate this journey,we recommend federal credit agencies to i)make AI adoption a priority,ii)develop a strategy,iii)set up the enablers,iv)develop and implement a robust roadmap,and v)iterate and monitor AI post-implementation for continued improvement.These steps are shown in Figure 7.Figure 7:AI adoption journey in federal credit 3.1.Make AI a priority Lets break each step down.First,to make AI adoption a priority,federal credit organizations should empower an initial AI organization and establish a budget.For any AI effort to get off the ground,the commitment of top-level leaders,organization-wide communication,and dedication of appropriate resources are absolute necessities.These are requirements irrespective of the maturity of an AI program but are of particular importance in the early stages.Some use cases may require effort by a subunit of a federal credit program while other use cases may require an agencywide approach or an approach that includes multiple agencies.At whatever level,it is likely that AI portfolio of initiatives will demand creation of a working group in which all relevant parts of an agency are represented and strong contractor support to help the agency develop and roll out the intended AI application.As for budget,federal credit program leaders should incorporate AI needs into their annual administrative budget requests as soon as possible.These budgets should include funds for 1PrioritizeAdoption2DevelopStrategy3Set UpEnablers4ImplementRoadmap5Monitor 31 market research,infrastructure and licensing,hiring AI subject matter experts(SMEs),contractor and vendor support,programming,and training.Additional budget will be needed for infrastructure and talent development.3.2.Develop an AI strategy After AI has been prioritized with top management,it is vitally important that each federal credit organization develop an AI strategy that defines its vision,goals,and objectives.Key initiatives should be identified,along with critical enablers and a robust roadmap for implementation of the strategy.One place to start the strategizing is by engaging key stakeholders.In addition,internal interviews and market research should be conducted.Through these discussions,high value initiatives,and key gaps,and areas that need additional resources and upskilling should be identified.Typical maturity assessment dimensions include technology and tools,data management,talent and skills,governance and compliance,culture and change management,and risk management.As a part of AI adoption strategy development,an inventory of existing,planned and potential use cases should be collected and evaluated by their potential impact,feasibility,cost,risks and return of investment.Within a federal credit organization,this could entail exploring AI use cases described in Section 2 of this paper.Federal credit agencies can expand their market research and experiment with agency-specific application and vendor AI tools to increase their conviction and understand the power of AI.A detailed assessment of vendor solutions versus in-house development options is also needed before making decisions.Moreover,agencies should look to each other for support,creating collaboration and synergy opportunities as well as places where costs can be minimized by developing solutions jointly.Federal Credit case example 4 Call for AI use cases Federal Student Aid(FSA)issued a request for information in September 2025 on technology modernization,where they are also seeking input from industry partners on best practices,proven tools,and innovative approaches to responsibly implement AI-driven personalization within a secure and compliant federal environment.18 These use cases will provide insight on how to best leverage AI to optimize service delivery,enhance customer experience,and support partner institutions;and include risk modelling to identify borrowers who may be at risk of delinquency and offering early outreach or relevant resources.They envision leveraging AI technologies to deliver tailored communications and support experiences based on individual user profiles,behavior,and context.This could include but is not limited to:18 SAM.gov FSA Technology Modernization Initiative RFI Response-Round 1,Question 27 32 Recommend relevant next steps in a students financial aid journey,such as reminders to complete FAFSA corrections,submit required documentation,or enroll in a repayment plan Personalize content and language based on a users circumstances a first-time FAFSA applicant versus a borrower in the process of repaying a loan and preferred communication channels Improve accessibility and user satisfaction by enabling conversational AI and natural language processing to power intelligent virtual assistants that can guide users through complex processes in real time Detect and respond proactively to potential issues or risks such as identifying borrowers who may be at risk of delinquency and offering early outreach or relevant resources.FSA is particularly interested in AI capabilities that can self-service,reduce information gaps,and improve outcomes all while maintaining rigorous standards for data privacy and responsible AI in automated decision-making,ultimately delivering measurable benefits:FSAs ongoing AI investments include:Tool catalogue Description Aidan Virtual Assistant An AI-powered chatbot deployed on StudentAid.gov that uses Azure OpenAI services.Aidan effectively deflects high-volume inquiries and provides timely guidance on FAFSA completion and loan repayment Agentic IVR Pilot Conversational AI prototype using AWS Bedrocks Claude 3.0 Haiku,improving call center experiences Enterprise Data Management&Analytics Platform Services(EDMAPS)Foundation for data aggregation enabling advanced analytics such as predictive modeling,reporting,visualization)to inform decisions Risk&Fraud Detection Tools(Exploratory)AI-driven anomaly detection to enhance data integrity and prevent fraud across application and servicing processes Strategy development and gap assessment will be key inputs to develop an implementation roadmap for the enterprise and use case specific initiatives.Implementation resources could then be prioritized to fulfil the gaps and for initiatives that offer a higher expected return on investment,allowing for the greatest impact within a limited budget and ease such as the ones for high-volume,simple reasoning processes.It is also important to prioritize initiatives that are easier and quicker to implement to achieve rapid success through such things as basic research-intensive tasks or the deployment of vendor AI tools that require no development effort.In high level interviews conducted for this report,we have identified several such initiatives including:Having an AI assistant or co-pilot for federal credit analysts to conduct research,review documents and draft reports,Applying machine learning techniques to a combination of structured and unstructured data for better risk assessment and real-time portfolio monitoring,Reducing fraud through early identification and better prevention,33 More accurate estimation of credit subsidies also with less hassle,and Using chatbots to support loan applicants.Figure 8 presents an illustration with 22 assumed AI initiatives for an agency.For illustrative purposes,these initiatives are mapped onto a matrix that classifies them into low,medium,and high levels for both impact and difficulty.Impact is assessed in terms of cost efficiency,effectiveness,and the introduction of new capabilities,or revenue uplift these are quite challenging to estimate and require iteration.In-house development efforts and vendor tool implementation options are also depicted along with opportunities for collaboration across federal credit agencies.Figure 8.Illustration of the output of a strategic review of potential AI initiatives In-house development efforts require special attention to design a target solution and subsequently build proof of concepts with the support of internal and contracted experts before piloting and scaling up.For instance,within a federal credit organization,this could mean working closely with loan officers to create mock-ups of an AI methodology and/or tool that predicts borrower default risk based on various indicators.This approach allows for a user-friendly experience and makes sure the end user is comfortable with the new tool.Building a proof-of-concept and creating mock-ups in a safe development environment allow teams to test assumptions and refine the AI solution.Pilot programs follow the proof-of-concept stage and are typically run with a subset of employees.For instance,a program could be developed around the evaluation of how a new AI tool can streamline a part of the loan application review process,such as gauging ImpactASSESSMENT OF AI INITIATIVESDifficulty of Development and ImplementationHighHighMediumMediumLowLowIn-house development$Cost efficiency$ Effectiveness,totally new capability Revenue upliftVendor tool implementationJoint development with another Federal Credit Agency$ $ $34 eligibility and completeness of an application.Once a minimum viable product is created,feedback sessions ensure that stakeholders are engaged in product development and ownership,and the final product would meet user needs.3.3.Set up key enablers To successfully execute the AI adoption framework,federal credit organizations must invest in four key foundational enablers i)governance,ii)technology and data,iii)monitoring,testing,and controls,and iv)talent and organization.In what follows,we discuss each enabler in detail,providing practical guidance from the private sector to help federal credit organizations unlock the full potential of AI while managing risk and adhering to best practices.3.3.a.Governance AI regulation,governance,and risk management are still at an early stage of maturity and evolving rapidly.Leading firms are introducing both“top-down”governance frameworks and“bottom-up”risk assessment of specific AI use cases.Figure 9.Main building blocks of AI governance Top-down governance is based on five main building blocks,as shown in Figure 9:Organizational strategy:The core of governance is an overall approach to AI strategy,the AI operating model,and organizational alignment.This incorporates organizational goals for AI,the main technology choices,areas for priority focus,budgeting and resourcing,and where and how decisions about AI adoption are made.Rules of the road:This sets the boundaries for AI deployment,consistent with laws,regulation,ethical considerations and the organizations risk appetite.Risk framework:This establishes the core concepts for AI risk management and identifies roles and processes for different functional areas,such as IT,data management,AI developers/business owners,and second line of defense functions,35 including risk management and compliance,legal,model risk management(MRM),and third party/vendor risk management Risk assessment:This develops the principles for assessing,approving and monitoring specific use cases.Organizational reinforcement:This focuses on broader organizational requirements for large-scale adoption,including education,training,resourcing,upskilling,and communication.Bottom-up risk assessment is derived from the same components and principles introduced above,and focuses on:Developing evaluation criteria for types of models and families of use cases Establishing tiered review,approval,and monitoring procedures reflecting varying degrees of risk of models and use cases Applying evaluation criteria and tiered assessment to assess,review,and approve specific AI applications Developing specific controls,safeguards,and monitoring procedures(such as human-in-the-loop)to ensure responsible deployment of a particular use case Monitoring models post-deployment and dynamically updating controls as warranted Both top-down governance frameworks and bottom-up assessment approaches need to be aligned with the state of organizational maturity.Figure 10 outlines governance solutions based on increasing maturity of the AI program from“aware”to“experimental”to“operational”to“mature”to“advanced”.The requirements for the“experimental”stage of AI adoption within an organization where models are being tested but still run alongside existing processes are much lighter than what is needed to support full-scale operational rollout.Leading organizations are beginning to define what“governance at scale”requires.Organizations often centralize their AI efforts initially sometimes establishing a GenAI center of excellence.As maturity increases,we see transitioning to a federated or hub-and-spoke model,where different functions and teams can independently develop and manage AI use cases with appropriate oversight.Also,importantly,a core pillar of AI governance is AI model evaluation,third-party assessment for vendor tools and model risk management.From original design,calibration and development of LLMs to various guardrails and controls built on top of LLMs to prompts used for the use cases,AI applications span multiple functional layers,each introducing complex and novel risks requiring tailored evaluations.These technical evaluations include automated and human tests of such model features as its accuracy,clarity,stability,completeness,confidence,bias,toxicity,robustness,sensitivity,and sycophancy.They are quite challenging as they require development of new capabilities and articulation of an agencys risk appetite to test,evaluate and validate AI outputs.36 Governance structures must also set boundaries around who can access models,what controls,guardrails and necessary training are in place,how AI content created is grounded,and how AI policies will be enforced.Controls are needed to identify and mitigate risks,and guardrails must be in place to detect anomalies and filter undesirable outputs such as hate speech and AI hallucinations.AI-output must be grounded and connected to verifiable data sources through confirmed source citations and confidence scores based on factual accuracy and transparency.Finally,there must also be policies for safe,secure,and trusted use of AI.Figure 10.AI Governance solutions need to be aligned with maturity 3.3.b.Technology and data Technology often represents the first barrier organizations face in adoption of AI.Secure access to LLMs and structured sandboxes for experimentation are essential to adopt AI.A well-designed tech stack including a sandbox allows AI developers to test use cases in a controlled environment,minimizing exposure to sensitive data and reducing risk.For security reasons,hosting open-source LLMs in-house is becoming a preferred approach.However,this requires a secure infrastructure with appropriate firewalls and encryption protocols to protect data privacy.Such development work is often led by IT.As of the publication date of this white paper,federal agencies have evaluated and deployed a range of LLMs.These include OpenAIs GPT family including GPT-4,Anthropics Claude,Googles Gemini/PaLM family,Microsofts Azure-hosted OpenAI models,and AWS Bedrock-hosted AwareOrganizations become awareof Gen Al anditspotential benefits.Exploration of LLM technologies,and consideration ofhow Gen Al might be used withinthe organization.Focus at this stageis primarily on learning andunderstanding.ExperimentalOrganizations have a basic understanding of Gen Al,and begin conducting small-scale LLM use-case experiments.Usually involves applying Gen Al specific,isolated business problems,often with external help.Goal at this stageis to gain practical experience,and understand thetechnical stack requirements and business challenges involved in implementingGen Al.The operating model at this stage may bea funded“program”oranAl Factory setup with some core capabilities to effectively andsafely experiment.OperationalOrganizationshave successfully completed several LLM use-cases and arenowworkingto integrate the use into their regular operations.Focus on scalingup Al capabilities,upskilling,and investing in necessary resources(such as data science talent).Operating Modelmay evolve towards embedding of additional capabilities to the Al Factoryin a centralized or hub-and-spoke archetype.MatureOrganizations have fully integrated Al into their operations and strategy.They have developed strong Al capabilities,including a skilled team.robust data infrastructure,vendor integration and effective risk and governance mechanisms.Al initiatives are driven more and more by internal teams rather than external parties.The operating model might transition towards adistributed model to leverage collective intelligence and drive innovation.AdvancedAl is a core part ofthe organizations strategyand culture.Al initiatives are proactive,anticipating andshaping business trends and customer needs,rather than simply responding tothem.Organization continuously innovates,driven by Al,and exhibits agile and resilient characteristics in rapidly adapting Al strategy based on market dynamics.37 partner models,as well as open-source and self-hosted models and derivatives like Metas LLaMA variants,Falcon,BLOOM,Vicuna,and other fine-tuned private instances.In general,AI tools must be agnostic to LLM models where possible,allowing teams to deploy different models without vendor lock-in.This includes selecting appropriate hardware,such as high-performance servers,and cloud computing resources,which can handle the extensive data processing and storage requirements of AI applications.Additionally,institutions should prepare for potential price volatility,especially when models are priced based on usage volume.Detailed infrastructure requirements are out of scope of this white paper.In 2025,the United States governments AI infrastructure requirements,primarily driven by the National Institute of Standards and Technology(NIST)and accompanying policies,emphasize a comprehensive approach to AI that includes innovation,risk management,technical standards,and addressing vendor-related challenges.For content on these developments,one can consult the NISTs official resources.Strong data foundations and protections are equally important.Effective data management and confidentiality measures,access controls and auditing mechanisms are needed to maintain data quality,integrity,and security.Government agencies often deal with fragmented,inconsistent,or outdated data,which can hinder effective AI implementation.Modernizing data infrastructure is crucial.Tactically,government organizations may want to explore AI use cases that rely on available private data.Document-level access controls are crucial for respecting individual-level permissions,an important consideration for sensitive federal datasets.Additionally,organizations should develop and enforce data retention and deletion policies to facilitate the deletion of outdated data or confidential prompts.3.3.c.Monitoring,Testing,and Controls Organizations deploying AI must ensure that the technology performs safely and ethically over time.Strong monitoring and control mechanisms should be implemented to track AI outputs,compliance,performance drift,and bias.Establishing feedback loops is important to detect when AI systems produce anomalous results or violate policy thresholds.Controls should be tailored to the organizations maturity.In less mature environments,AI systems should only operate in tightly monitored contexts with clear override capabilities.In more advanced settings,automated alerts and model performance dashboards can support ongoing risk management at scale.Across the board,guardrails can monitor AI output via a variety of tasks,such as semantic validation,detection of personally identifiable information leaks,and identification of harmful content.Teams must differentiate between testing models to create them versus testing models to validate use cases.Testing must be structured not only to prove technical accuracy but also to demonstrate policy compliance and minimize third-party risk.Risk-based testing and monitoring standards should be implemented.For example,a federal credit organization might implement tiered risk assessments that classify AI models used in underwriting based on their risk levels,with higher-risk models undergoing more stringent validation processes.38 3.3.d.Talent and Organization Building and sustaining AI capabilities requires cross-functional teams with technical,analytical,and operational expertise,in addition to a strong group of data scientists.Organizations should focus on training and upskilling their broader workforce not just AI experts to interact effectively with AI tools.Education,upskilling employees and change management are crucial for evolving business processes to incorporate AI.Offering training sessions and interactive workshops that cover AI fundamentals,data literacy,prompt writing guidance,grounding,and specific tools will empower federal credit employees to leverage AI effectively in their roles.Practical,hands-on training opportunities will reinforce concepts and build confidence in using new tools.In these training programs,learning from examples would increase the rate of adoption and drive transformation.Also importantly,encouraging a culture of ongoing education through access to online courses,workshops,and industry conferences will keep staff informed about the latest developments in AI.Additionally,organizations can foster a culture of experimentation by supporting joint teams from business and technical backgrounds in developing small AI pilot projects to experiment with the technology in a controlled setting.To facilitate a smooth transition,it is important to communicate the vision and benefits of AI to all employees,addressing any concerns or resistance and emphasizing the benefits of human-machine collaboration.Change management strategies should include mechanisms for monitoring the impact of AI on service delivery and compliance,so that the transition supports the overall mission of the federal credit organization with minimal disruption.3.4.Develop and implement a robust roadmap Based on use case opportunity evaluation,the needs to implement or improve the enablers,and available resources,a roadmap for the whole portfolio of initiatives can then be developed.Such a roadmap serves as a guide for implementation and provides clarity on objectives,timelines,dependencies,and resource allocation,ensuring alignment with organizational goals.Successful execution of the roadmap and associated processes demands professional project management applied to both program and initiative levels.Applying agile methodologies enhances flexibility,allowing teams to adapt seamlessly to changes.Incorporating feedback into the roadmap is crucial for fostering continuous learning and improvement.By soliciting insights from diverse stakeholders and tracking performance through key performance indicators,organizations can refine their strategies in response to real-time data and shifts in technology.39 Clear communication with stakeholders throughout the process is important for encouraging engagement and collaboration.Additionally,careful resource allocation and robust risk management practices significantly increase the chances of successful outcomes.By prioritizing these essential elements,organizations can successfully leverage AI initiatives,promoting sustainable growth and innovation while remaining aligned with their broader strategic objectives.Figure 11:Portfolio of initiatives 3.5.Continue to iterate and monitor AI program Finally,federal credit organizations should continuously review the performance of their AI tools against defined objectives.By gathering feedback from users and analyzing outcomes,organizations can gain valuable insights to refine their strategies and enhance future or other AI projects.Additionally,providing ongoing training for employees and encouraging discussions about AI within teams and senior management will empower staff to embrace new tools and contribute to the AI journey.In parallel,federal credit organizations should monitor the developments in the private sector to learn,assess and adapt best practices as technology develops and evolves.Checklist for action by federal credit Agencies We expect AI adoption to continue at a ground-breaking pace globally,and its important for federal credit organizations to act now and begin or accelerate their AI journey.There are number of immediate steps agencies can take to quickly ramp up AI adoption.In closing,we provide a generalized checklist for action:40 1.Prioritize AI Adoption Leadership commitment:Secure top-level support for AI initiatives.Establish a dedicated team:Form an AI organization with a working group responsible for steering adoption and strong contractor support.Initial budget allocation:Secure budget as soon as possible for market research,strategy development,infrastructure,licensing,planning,talent acquisition,contractors and necessary training 2.Develop an AI Strategy Formulate an AI adoption strategy with long-term vision,goals and objectives,may it be experimental or directly targeting a transformational change Engage stakeholders:Conduct interviews and gather insights from key stakeholders Conduct market research:Carry out market research on AI adoption in the private sector Inventory use cases:Document existing and potential AI use cases,evaluating them for impact and feasibility.Embrace profession-specific AI approaches and tools in the market,Assess in-house development options against vendor solutions.Explore federal credit wide collaboration and synergies across agencies,Identify high-value initiatives:Assess and prioritize initiatives based on their potential impact,feasibility,budget,and return on investment Conduct maturity assessment:Evaluate technological capability,data management,talent,governance,and risk management 3.Set Up Key Enablers Governance framework:Establish organizational strategy and risk management frameworks aligned with maturity and ambition of the AI adoption Define rules for AI deployment in line with legal and ethical considerations Develop AI related policies and guidelines Implement a tiered review and approval process for AI applications Technology and Data Infrastructure:Ensure safe,secure,controlled and trusted access to AI tools Develop strong data management practices,including access controls and data retention policies Transition toward modernized data management practices,ensuring integrity and compliance with federal standards Monitoring,Testing,and Controls:Integrate comprehensive testing routines to assess both the technical and ethical aspects of AI outputs Implement controls to manage AI output and ensure compliance Establish feedback mechanisms to track performance and monitor for anomalies 41 Talent and Skill Development:Focus on cross-functional teams with diverse skill sets Provide hands-on training on AI tools and prompt writing guidance Foster a culture of continuous learning and experimentation Proactively manage the change 4.Develop and Implement a Robust Roadmap Create an Implementation Plan:Develop a roadmap that prioritizes initiatives and includes timelines and milestones Feedback Loop:Ensure mechanisms are in place for ongoing learning and refinement of AI strategies 5.Iterate and Monitor Post-Implementation Performance Review:Continuously review AI tools against defined objectives and gather user feedback Training and Support:Provide ongoing education and support to employees Benchmarking:Monitor private sector best practices to adapt and enhance AI governance and strategies AI solutions go beyond productivity gains to enhance quality of the work products and drive innovation throughout the credit lifecycle.By following the above-mentioned five steps,we believe federal credit organizations can effectively navigate the complexities of AI adoption,harness its transformative potential,and position themselves for success in an increasingly AI-powered landscape.42 4.Appendix-AI Supremacy Innovation and adoption in AI are continuous processes.The dynamic state of AI is chronicled by numerous publications.For example,Stanford HAI AI Index 2025 Report captures and summarizes the evolving state of the field for a broad audience.19 This section provides essential background information,definitions,and context about predictive,generative,and agentic AI,as well as their use cases in financial services,associated risks,and regulation by the time of the publication of this report.4.1.Definitions and growing impact The Merriam Webster Dictionary defines artificial intelligence(AI)as the capability of computer systems or algorithms to imitate intelligent human behavior.20 This definition is broad,encompassing various types of AI,each with distinct capabilities,limitations,and applications.Among these,predictive,generative,and agentic AI are gaining prominence.Predictive AI focuses on analyzing data to forecast future outcomes,GenAI creates new content based on learned patterns.And finally,agentic AI refers to systems that can operate autonomously,perform a variety of operations,and make decisions in real-time giving AI agency rather than restricting it to tool status.Understanding these types of AI and their growing impact is foundational for federal credit organizations looking to leverage technology for improved efficiency and innovation.4.1.a.Predictive AI Predictive AI refers to systems that analyze historical data to make informed predictions about future events or outcomes.By employing statistical techniques,machine learning algorithms,and data mining methods,predictive AI identifies patterns and trends within large datasets.Predictive AI has practical applications in everyday life and is seamlessly integrated into the technology around us,enriching user experiences and optimizing decision-making processes.Predictive AI is also commonly used by businesses to anticipate customer behavior,sales,prices,or demand for products,allowing them to adjust their strategies accordingly.For example,it suggests next songs based on a users existing playlist,recommends credit cards based on financial history,and personalizes internet search results based on search history.Predictive AI can also be used by the federal credit sector to improve data-driven decisions,which has historically been a hallmark of the federal credit industry.Now,predictive AI allows for advanced methodologies to be built and tap into much larger and unstructured datasets with alternative content compared to historical information.One example includes using predictive AI to improve the estimation of probability of default of loans by applying machine learning algorithms and incorporating alternative content in unstructured data such as news,social media,court files,transaction and payments data,pared to 19 https:/hai.stanford.edu/ai-index/2025-ai-index-report 20 ARTIFICIAL INTELLIGENCE Definition&Meaning-Merriam-Webster 43 traditional datasets such as financial information about the borrower,qualitative information about management and industry outlook.Advanced modeling and alternative data with new and orthogonal information help us to include previously not captured risk factors,interaction across factors,and non-linear phenomena in the probability of default models,and automatically optimize data-driven segmentation.4.1.b.GenAI While predictive AI excels in analyzing existing data to forecast user preferences and enhance personalization,GenAI takes this a step further by not only understanding these patterns but also creating entirely new content and experiences based on that knowledge.This technology generates text,images,audio,and even video content that closely resembles human-created outputs.One of the most notable applications of GenAI is in natural language processing(NLP),where models like OpenAIs ChatGPT can generate coherent and contextually relevant text based on prompts provided by users.This capability opens new possibilities for content creation,marketing,and customer service,as businesses can leverage GenAI to produce reports,articles,advertisements,management reports,and responses to customer inquiries efficiently.In addition to text generation,GenAI is also used in graphic design,videos,speech,and art,enabling the creation of unique images and visual content.Furthermore,it has applications in fields such as drug discovery,where generative models can propose new molecular structures based on existing compounds,accelerating the research and development process.In the federal credit sector,professionals can leverage GenAI across the investment process to enhance and speed up analysis,synthesis,reasoning,communication,and reporting and information management processes.More use cases are detailed in Section 2,but one example includes providing loan appli
2025-11-03
59页
5星级
Issue BriefOctober 2025The Mechanisms of AI HarmLessons Learned from AI IncidentsAuthorMia HoffmannT.
2025-11-03
36页
5星级
How to Create ChatGPT Apps:A DevelopersGuide(2025)By Adrien Laurent,CEO at IntuitionLabs 10/1/2025 3.
2025-11-03
15页
5星级
State of Tech ExitsGlobal H1 2025 Global data and analysis on tech exit activity and market liquidityAction is better.Action is better.We track and compare the worlds companies.Our AI agents keep you a step ahead of competitors and disruption.Trusted by the worlds smartest companies:Every big tech company.Every top professional services firm.26 of the top 30 banks.All ten Fortune 10 companies.Put our AI agents to work today.IIIIIIStart free trialTL;DRTL;DRYour rundown ontech exits in H125122tech companies went public The global tech IPO market has remained muted during the first half of 2025,with 122 tech companies going public,in line with the numbers from 2024.See the data$2Traised in equity funding by private tech companiesPrivate tech companies are staying private longer and now have more capital than ever to do so.Over$2T in cumulative equity funding has poured into private tech markets to date,with 90%raised in just the last decade.That funding has enabled companies to keep scaling before tapping the public markets.See the data 7thconsecutive quarter of YoY growth in secondariesThe last 7 quarters have seen YoY growth in secondary transaction activity among VC-backed companies,with no signs of slowing down.As tech companies stay private longer and valuations continue to climb,secondaries are playing a growing role in providing liquidity.See the data 1new exit model emerges amidst AI talent warThe intensifying race for AI talent combined with tight regulation aredriving a new wave of unconventional exits in the tech ecosystem,bypassing traditional M&A.Reverse acqui-hires where acquirers buy the team(fully or partially)and license the technology in particular are becoming increasingly popular.See the data$100M deals drive tech M&A momentumTech M&A activity has remained stubbornly flat since Q423,stagnating at just over 2,000 transactions per quarter,with AI capturing a growing share.Despite flat volume,this year is shaping up to be a record year in terms of M&A deal value,driven by an increase in the number of$100M acquisitions.See the data AI and$100M deals drive tech M&A momentumWe project Q325 growth to be flat,with 2,040 dealsTL;DRTL;DR State of Tech Exits 2025 CB Insights.Data as of 8/20/2025.Signs point to tech IPO market rebound in H225Recent activity such as Figmas and Bullishs IPOs signals things may be picking upTL;DRTL;DR State of Tech Exits 2025 CB Insights.Data as of 8/20/2025.Excludes SPACs.Private tech markets top$2T in equity funding90%of it was raised in just the last decadeTL;DRTL;DR State of Tech Exits 2025 CB Insights.Data as of 6/30/2025.Secondaries get bigger and pricierAverage discounts in secondary markets have compressed to just 13low last-round valuationsTL;DRTL;DR State of Tech Exits 2025 CB Insights.Source:EquityZenNew exit models emerge amidst the AI talent warReverse acqui-hires are gaining momentum among big tech companiesTL;DRTL;DR State of Tech Exits 2025 CB Insights.Data as of 8/13/2025.ContentsContentsGeneral trends10Private tech market trends12New exit models14Secondaries trends16M&A trends21AI M&A trends36Tech trends43State of Tech Exits10General trendsTech companies stay private for longer,going public 16 years after being founded compared to 12 years in 201511State of Tech Exits|General Trends|H1 2025 2025 CB Insights.Data as of 6/30/2025Note:excludes SPACs.Y-axis begins at 5 years12.211.913.911.813.414.013.715.215.415.715.9years20152016201720182019202020212022202320242025 YTDAverage age at IPO4 yearssince 2015But they have no problem finding money,topping$2T in equity funding with 90%of it raised in the last 10 years12State of Tech Exits|General Trends|H1 2025 2025 CB Insights.Data as of 6/30/2025$0.0T$0.5T$1.0T$1.5T$2.0TQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320152016201720182019202020212022202320242025$2.0TCumulative equity funding for private tech companies$0.2T20152016201720182019202020212022202320242025 YTDSeedWhich year saw the largest round by stage?Private tech companies are raising ever-larger rounds across stages13State of Tech Exits|General Trends|H1 2025 2025 CB Insights.Data as of 6/30/2025Series ASeries BSeries CSeries DSeries E 202520252024201820252024$2B$3B$6B$5B$10B2022$3BReverse acqui-hires gain traction as a new exit models amid AI talent war and regulatory pressures14State of Tech Exits|General Trends|H1 2025DateJul 2025Jun 2025Aug 2024Aug 2024Jun 2024Mar 2024Valuation pre-acquisition$1.3B$13.8B$1.0B$0.6B$1.0B$4.0BDeal typeReverse acqui-hireKey talent acquisition corporate minorityReverse acqui-hireReverse acqui-hireReverse acqui-hireReverse acqui-hireTransaction detailsGoogle licensed technology from AI coding startup Windsurf for$2.4B and hired key personnel including CEO Varun Mohan and co-founder Douglas Chen to join DeepMind division.Meta took a 49%minority stake in Scale AI for$14.8B,valuing the data labeling company at over$29B post-money.Scale founder Alexandr Wang joined Meta to lead a new superintelligence division.Google rehired Character.AIco-founders Noam Shazeerand Daniel De Freitas along with some Character.AIstaff,integrating them into its Google DeepMind team.As part of the non-exclusive agreement,Google will license Character.AIstechnology for$2.7B.Amazon hired Covariantsfounders along with a quarter or so of the startups employees,while paying a rumored$440M fee for a non-exclusive license to Covariants robotic foundation models.Amazon is licensing Adepts agent technology,multimodal models,and datasets for a rumored fee of$330M.Some Adept team members will join Amazons AGI organization.Microsoft brought in Inflection AIs team and licensed its technology for a$650M fee,after an initial investment.acquired byacquired byacquired byacquired byacquired byacquired by 2025 CB Insights.Data as of 8/13/2025Top LLM developers have recently been opting for acqui-hires over traditional acquisitions15State of Tech Exits|General Trends|H1 2025 2025 CB Insights.Source OpenAIs Acquisition Insights;Anthropics Acquisition InsightsYoY growth rate of secondary transactions for VC-backed companiesSecondary transactions are growing fast,providing liquidity to both long-term investors and employees without formal exit16State of Tech Exits|General Trends|H1 2025 2025 CB Insights.Data as of 6/30/2025-40%-20%0 000%Q121 Q221 Q321 Q421 Q122 Q222 Q322 Q422 Q123 Q223 Q323 Q423 Q124 Q224 Q324 Q424 Q125 Q225Investor interest by company age(as a%of overall investor interest)Investor interest by company age(as a%of overall investor interest)Investor interest in 2025MoM growth rateInvestor interest in 2025MoM growth rateInvestor demand for secondary stakes in private companies is rising,driven by interest in younger companies17State of Tech Exits|General Trends|H1 2025Source:EquityZen-13%-6%-382V%Jan25Feb25Mar25Apr25May25Jun25Issuers:=5 years old6-10 years old10 years old0 0Pp0%Average trading discount rate to last-round valuationAverage trading discount rate to last-round valuationSecondary trading discount/premium distributionBased on deal launch dateSecondary trading discount/premium distributionBased on deal launch dateHigher demand is leading to lesser discounts to last-round valuation18State of Tech Exits|General Trends|H1 2025Source:EquityZenPremiumDiscount0 0Pp0%-35%-47%-51%-55%-49%-29%-33%-32%-28%-13%-60%-50%-40%-30%-20%-10%0%Q123 Q223 Q323 Q423 Q124 Q224 Q324 Q424 Q125 Q225 The 10 most popular companies on EquityZens platform in Q225The 10 most popular companies on EquityZens platform in Q225Most popular industries for secondary transactionsAs of Q225Most popular industries for secondary transactionsAs of Q225In secondaries:AI dominates,SaaS is down,crypto makes a comeback19State of Tech Exits|General Trends|H1 2025Source:EquityZen01234567Q124Q224Q324Q424Q125Q22587654321RankArtificial intelligenceInformation technologyFintechAerospaceNational securityCryptoManufacturingSaaS12345789106AerospaceArtificial intelligenceArtificial intelligence national securityArtificial intelligenceArtificial intelligenceCryptoArtificial intelligenceArtificial intelligenceArtificial intelligenceArtificial intelligenceTop Mosaic score companies have gained in popularity among secondary investors20State of Tech Exits|General Trends|H1 2025 2025 CB Insights.Source:EquityZen;CB Insights;data as of 8/13/2025.Note:Mosaic scores measure a private companys health,with a score of 510 or above being considered highCompanies that grew the most in popularity on EquityZens platform in Q225,ranked by Mosaic score12345678910889 52 YoY850 155 YoY887 26 YoY845-3 YoY882 163 YoY823 213 YoY878 3 YoY665 90 YoY867 127 YoYN/AN/A21M&A trendsState of Tech Exits|M&A Trends|H1 2025Tech companies are getting acquired earlier,and at a higher priceStats for the typical acquired company in 2025 YTD22Median valuation $8MAverage yearsto exit-3.4 yearsMedian headcount at acquisition 2 employees$27M$35Min 2024in 2025 YTD22.7 years19.3 yearsin 2024in 2025 YTD31employees33employeesin 2024in 2025 YTD 2025 CB Insights.Data as of 6/30/20252025 is shaping up to be a record year in terms of M&A deal value,despite flat volumeGlobal tech M&A deal volume and aggregate disclosed target company valuation annual,20212025 YTD23State of Tech Exits|M&A Trends|H1 2025 2025 CB Insights.Data as of 6/30/2025$715.1B$694.1B$498.6B$403.7BAggregate valuationso far$434.0B9,8519,0397,7748,201Deals so far4,07120212022202320242025 YTDQuarterly M&A deal value maintained momentum in Q225,reaching levels not seen since 2023Global tech M&A deal volume and aggregate disclosed target company valuation quarterly,20212025 YTD24State of Tech Exits|M&A Trends|H1 2025 2025 CB Insights.Data as of 6/30/2025$156.5B$190.2B$189.2B$179.1B$309.7B$118.9B$140.0B$125.6B$99.0B$95.7B$64.8B$239.1B$74.1B$117.3B$102.7B$109.6B$211.3BAggregatevaluation$222.8B2,3352,3672,4552,6942,6812,2972,0751,9862,0381,9121,7882,0362,0112,0162,0892,0852,021Deals2,050Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q220212022202320242025Large deals see rebound from 2023 lowsAnnual number of$100M tech M&A transactions,20212025 YTD25State of Tech Exits|M&A Trends|H1 2025 2025 CB Insights.Data as of 6/30/2025Percent of total deal share6.2%3.9%3.4%3.8%4.7a5352266314190so far20212022202320242025 YTDxAISpectrumAlphabetGlobal PaymentsEmersonSiemensSilver LakeBlackstone|Vista Equity PartnersBrookfield Infrastructure PartnersCVC Capital Partners|Abu Dhabi Investment Authority|Nordic CapitalThe AI boom drives 4 of the largest tech acquisitions so far this year2025 YTDs top 10 largest acquisitions by disclosed valuation26 2025 CB Insights.Data as of 6/30/2025Note:exclude pending acquisitionsAcquirersState of Tech Exits|M&A Trends|H1 2025Acquired companyValuation$6.9B$7.0B$8.4B$8.8B$10.0B$16.8B$22.7B$32.0B$34.5B$45.0B27State of Tech Exits|M&A Trends|H1 2025 2025 CB Insights.Data as of 6/30/2025-36%vs.Q221$54M$56M$55M$46M$40M$40M$39M$30M$24M$29M$31M$40M$17M$34M$25M$30M$33MMedianvaluation$36MQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q220212022202320242025Median valuations are on an upward trajectory,remain below 2021 record levelsQuarterly median valuation of acquired companies,20212025 YTD4 out of 10 companies are still sold at a discount to their last available valuation28State of Tech Exits|M&A Trends|H1 2025 2025 CB Insights.Data as of 6/30/2025Q121Q221Q321Q421Q122Q222Q322Q422Q123Q223Q323Q423Q124Q224Q324Q424Q125Q22535%-13%-2%-2%-10)!1AHRUD8A5UPBRU290)!%Median premium or discount%of transactions made at a discount80%of M&A deals involved companies with 100 employees or lessQuarterly tech M&A deal share by number of employees,20212025 YTD29State of Tech Exits|M&A Trends|H1 2025 2025 CB Insights.Data as of 6/30/202564abbgfdfiheegdfdf%7%7%7%8%6%6%6%6%5%6%6%6%5%6%6%7%7%3%3%4%3%4%3%4%4%4%3%4%4%3%4%3%4%3%9%9%8%9%Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q22021202220232024202511%|200 employees3%|151-2006%|101-15016%|51-10064%|50%)of a private companys shares.The acquired company gets added to the investors investment portfolio with the goal of increasing the value of the company over time and eventually selling it off for a profit.Acquisition(Talent)or Acquihire:The practice of acquiring a private company for the sole purpose of recruiting the staff rather than the product/service.Asset Sale:Most or all assets of a private company are purchased due to bankruptcy/administration/insolvency.Corporate Majority:Private or public company purchases a majority(50%)of a private companys shares.Leveraged Buyout:A transaction where a company is acquired through borrowed funding(bonds or loans)to meetcosts.Management Buyout:A transaction where the companys management purchases the assets and operations of the company they manage.Merger:An agreement between two private companies combining their businesses.Take Private:A private equity firm,a consortium of private equity firms,or another investor purchases the stock of a publicly traded company.The transaction converts the acquired public company to a private company.What is excluded:We exclude Reverse Mergers,such as SPACs.Anyone elseremember SPACS?What a great grift that was.49State of Tech Exits
2025-11-03
50页
5星级
State of AIGlobal|Q3 2025Global data and analysis on dealmaking,funding,and exits by private market AI companiesState of AI2 2Action is better.We track and compare the worlds companies.Our AI agents keep you a step ahead of competitors and disruption.Trusted by the worlds smartest companies:Every big tech company.Every top professional services firm.26 of the top 30 banks.All ten Fortune 10 companies.Put our AI agents to work today.IIIIIIStart free trialState of AI3 3“Steal”our data.Download the underlying data in this report.If you create any analysis or visualizations with this data,send it our way and you could be featured in the CBI newsletter.us on social LinkedIn X(Twitter)InstagramIIIIDownload the raw dataTL;DR 2025 CB Insights|4Your rundown onAI in Q325172M&A deals,staying elevated.AI M&A activity remained near record highs with 172 deals in Q325,just behind Q225s 181 deals.Notably,three of the quarters 5 biggest acquisitions involved AI agent companies.Legacy enterprise software players are aggressively acquiring to fast-track their AI product roadmaps.See the data 7GEO deals.Generative engine optimization(GEO)broke into the most active tech markets this quarter with 7 deals.These platforms help brands improve their discoverability on AI-powered search tools like ChatGPTand Perplexity.With OpenAIrecently rolling out shopping features in ChatGPT,AI platforms are becoming critical new commerce channels.See the data$104.3M Valuation per employee.Lean AI teams are commanding extraordinary valuations.Humanoid robotics company Figure topped the charts at$104.3M per employee based on its$39B valuation.These premium multiples reflect investor confidence,but the real test will be whether these startups can hit their aggressive revenue targets in coming years.See the data-22crease in deal activity QoQ.Global AI startup deals dropped to 1,295 in Q325,yet funding stayed strong above$45B for the fourth straight quarter.Deal sizes have swelled,with the 2025 YTD average hitting$49.3M an 86%jump from last year.Investors are making more concentrated bets as they chase AI winners amid high infrastructure costs and fierce competition in model development.See the data Bigger bets are the new normal in AITL;DR State of AI 2025 CB Insights.Equity deals only.Data as of 10/1/2025IIITrack 43,000 AI companiesAI acquisitions hit all-time highs,led by agentic solutionsTL;DR State of AI 2025 CB Insights.Data as of 10/1/2025.*Indicates valuation at time of exitIISee high potential M&A targetsGenerative engine optimization(GEO)emerges among most active tech marketsTop markets by equity deal activity in Q325 TL;DR State of AI 2025 CB Insights.Note:CB Insights Commercial Maturity scores measures(on a 5 point scale)a companys ability to compete for customers or serve as a partner.Mosaic scores measure private company health and growth potential(out of 1,000).Data as of 10/1/2025IIIIIIIMonitor 1,500 tech marketsSmall teams,big valuationsHumanoid robotics developer Figure leads top deals in Q325 by valuation per employeeTL;DR State of AI 2025 CB Insights.Data as of 10/1/2025.ContentsGlobal Trends10Investment Trends11Headcount Trends34Unicorns41Exit Trends47Investors53Geographic Trends57US58Silicon Valley64New York68Los Angeles72Boston76Seattle80Austin84Miami88Philadelphia92Canada96Asia102China108India112Singapore116Japan120Europe124United Kingdom130Germany134France138Israel142LatAm146Brazil152Oceania1569State of AINEW!Global Trends10Investment Trends11Annual equity funding&dealsState of AI|Global Trends|Investment Trends$107.3B$71.8B$69.2B$108.0BFunding$158.9B5,4815,4505,2395,725Deals4,48001,0002,0003,0004,0005,0006,0007,0008,000$0.0B$20.0B$40.0B$60.0B$80.0B$100.0B$120.0B$140.0B$160.0B$180.0B20212022202320242025YTD 2025 CB Insights|12Quarterly equity funding&dealsState of AI|Global Trends|Investment Trends$25.3B$28.0B$27.2B$26.8B$26.2B$19.3B$14.0B$12.3B$26.5B$14.7B$14.6B$13.4B$17.0B$26.7B$18.0B$46.4B$62.4B$48.7BFunding$47.8B1,3431,2441,4771,4171,6841,3611,2621,1431,4131,2951,2941,2371,3651,3741,5511,4351,5161,669Deals1,29502004006008001,0001,2001,4001,6001,8002,000$0.0B$10.0B$20.0B$30.0B$40.0B$50.0B$60.0BQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|13State of AI|Global Trends|Investment TrendsUS$38.8B652 DealsAsia$2.9B297 DealsEurope$5.4B279 DealsCanada$0.4B21 DealsLatAm$0.1B19 DealsOceania$93M17 DealsAfrica$14M7 DealsAll Other Regions$28M3 DealsFunding&deals by global region in Q325 2025 CB Insights|14Percent of quarterly deals by global regionState of AI|Global Trends|Investment Trends45DEDEEECRIQFTPGGTT%US,50(1)%&0$#% !$! %Asia,23!# &% !#% !%Europe,22%LatAm,1nada,2rica,1%Oceania,1%All Other Regions,0%Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|15Percent of quarterly deals by investor groupState of AI|Global Trends|Investment Trends3554443344655541552%VC,34%Other,16%9%Angel,11%9%9%9%9%8%8%9%8%8%9%8%8%8%8%9%9%Corp,10%9%9%9%9%9%8%8%9%9%8%8%8%8%8%9%8%CVC,8%9%6%8%6%9%6%7%7%7%9%8%9%Incubator/Accelerator,8%7%9%9%9%8%8%9%6%8%7%7%7%8%7%7%8%7%Asset/Investment Management,8%6%7%7%6%6%6%6%Private Equity,5%Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|16Annual average&median deal sizeState of AI|Global Trends|Investment Trends$27.2M$18.0M$18.6M$26.5M$49.3M202120232025YTDAverage Deal Size$5.0M$4.0M$3.0M$3.5M$4.7M202120232025YTDMedian Deal Size 2025 CB Insights|17Annual median deal size by global regionState of AI|Global Trends|Investment Trends$7.0M$4.8M$5.2M202120232025YTDUS$6.0M$2.3M$3.4M202120232025YTDAsia$3.2M$2.9M$4.1M202120232025YTDEurope$1.1M$0.9M$2.1M202120232025YTDLatAm$2.1M$2.3M$5.4M202120232025YTDCanada$0.5M$1.8M$2.9M202120232025YTDAfrica$3.0M$2.2M$3.2M202120232025YTDOceania 2025 CB Insights|18Annual median deal size by investor groupState of AI|Global Trends|Investment Trends$5.0M$4.8M$4.5M202120232025YTDAngel$50.0M$15.0M$20.0M202120232025YTDAsset/Investment Management$20.0M$16.0M$17.0M202120232025YTDCVC$15.5M$9.0M$13.3M202120232025YTDCorp$42.0M$20.0M$26.2M202120232025YTDPrivate Equity$14.0M$8.0M$10.0M202120232025YTDVC 2025 CB Insights|19Quarterly funding&deals from mega-rounds(deals worth$100m )State of AI|Global Trends|Investment Trends$17.2B$16.3B$15.2B$14.6B$13.4B$7.9B$5.6B$5.1B$19.7B$7.1B$7.3B$5.8B$9.7B$18.1B$8.2B$36.3B$51.7B$36.6BFunding$37.0B697077717148272815363529294226426463Deals630102030405060708090100$0.0B$10.0B$20.0B$30.0B$40.0B$50.0BQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|20State of AI|Global Trends|Investment TrendsUS$32.3B48 DealsEurope$3.0B5 DealsAsia$1.5B8 DealsCanada$0.2B2 DealsMega-round funding&deals by global region in Q325 2025 CB Insights|21Quarterly mega-round deals by global regionState of AI|Global Trends|Investment Trends19324744US,484678Asia,83389Europe,5LatAm,0111Canada,211Oceania,0Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|22Quarterly mega-rounds as percent of fundingState of AI|Global Trends|Investment Trends68XVTQABtHPCWhFxu%Mega-rounds,772BDFIYX&RPWC2T%Non-mega-rounds,23%Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|23Global:Top equity deals in Q325State of AI|Global Trends|Investment TrendsCompanyRound AmountRoundDateRound Valuation Select InvestorsCountry1Anthropic$13.0BSeries F2025-09-02$183.0BFidelity Investments,ICONIQ Capital,Lightspeed Venture Partners,General Catalyst,Insight PartnersUnited States2OpenAI$8.3BPrivate Equity2025-08-01$300.0BDragoneer Investment Group,Andreessen Horowitz,Sequoia Capital,Altimeter Capital,Bossa InvestUnited States3Mistral AI$1.5BSeries C2025-09-08$13.2BASML,Andreessen Horowitz,DST Global,General Catalyst,Index VenturesFrance4Nscale$1.1BSeries B2025-09-17N/AAker,Sandton Capital Partners,Blue Owl Capital,Dell Technologies CapitalUnited Kingdom5Databricks$1.0BSeries K2025-08-19$100.0BAndreessen Horowitz,Insight Partners,MGX,Thrive Capital,WCM Investment ManagementUnited States5Figure$1.0BSeries C2025-09-16$39.0BParkway VC,Align Ventures,Intel Capital,Brookfield Asset Management,LG Technology VenturesUnited States7Groq$750MSeries E2025-09-17$6.9BDisruptive,BlackRock,Cisco,D1 Capital Partners,Neuberger BermanUnited States8Ramp$500MSeries E2025-07-30$22.5BICONIQ Growth,General Catalyst,Lightspeed Venture Partners,Sutter Hill Ventures,137 VenturesUnited States9Cognition$400MSeries C2025-08-14$10.2BFounders Fund,Bain Capital Ventures,8VC,Elad Gil,D1 Capital PartnersUnited States10Sierra$350MSeries D2025-09-04$10.0B GreenoaksUnited States 2025 CB Insights|24Annual percent of deals by deal stageState of AI|Global Trends|Investment Trends69rwvrly-stage,74%Mid-stage,12%8%7%5%6%Late-stage,8%5%7%7%6%Other,6 212022202320242025YTD 2025 CB Insights|25Regional deal share by deal stage in Q325State of AI|Global Trends|Investment Trends49HQp)(# rly-stageMid-stageLate-stageOtherUSAsiaEuropeLatAmCanadaAfricaOceaniaAll Other Regions 2025 CB Insights|26Annual median funding by deal stageState of AI|Global Trends|Investment Trends$2.8M$2.8M$2.1M$2.5M$3.4M202120232025YTDEarly-stage$35.0M$30.0M$23.8M$27.2M$36.0M202120232025YTDMid-stage$85.0M$55.8M$41.1M$40.0M$68.6M202120232025YTDLate-stage 2025 CB Insights|27Global:Top seed/angel deals in Q325State of AI|Global Trends|Investment TrendsCompanyRound AmountRoundDateSelect InvestorsCountry1TARS$125MSeed VC2025-07-08Meituan,Linear Venture,Xiang He Capital,C&D Emerging Investment,CIVCChina2Genesis AI$105MSeed VC2025-07-01Eclipse,Khosla Ventures,Bpifrance,HongShan,Eric SchmidtUnited States3Upscale AI$100MSeed VC2025-09-17Maverick Silicon,Mayfield,Celesta Capital,Cota Capital,MVP VenturesUnited States4Radical AI$55MSeed VC2025-07-20RTX Ventures,AlleyCorp,Eni Next,NVentures,noaUnited States5Obot AI$35MSeed VC2025-09-23Mayfield,Nexus Venture PartnersUnited States6Wonderful$34MSeed VC2025-07-02Index Ventures,Bessemer Venture Partners,Vine VenturesIsrael6Aiphoria$34MAngel2025-07-21Ratmir TimashevUnited Kingdom8Aurasell$30MSeed VC2025-08-26Next47,Menlo Ventures,Unusual VenturesUnited States9JoyIn$28MSeed VC2025-09-28Eastern Bell Capital,IDG CapitalChina9Xingyuanzhi Robotics$28MSeed VC2025-09-10CAS Star,China ICV Investment,Hillhouse Capital Management,Oriza Equity Investment,Zhiyuan RobotChina 2025 CB Insights|28Global:Top Series A deals in Q325State of AI|Global Trends|Investment TrendsCompanyRound AmountRoundDateSelect InvestorsCountry1FieldAI$314MSeries A2025-08-20Bezos Expeditions,Prysm Capital,Temasek,BHP VenturesUnited States2Lila Sciences$235MSeries A2025-09-15Braidwell,Collective Global,General Catalyst,Ark Ventures,Flagship PioneeringUnited States3Lovable$200MSeries A2025-07-17Accel,Creandum,Visionaries Club,byFounders,HummingbirdUnited States3Periodic Labs$200MSeries A2025-08-08Andreessen HorowitzUnited States5UltraGreen$188MSeries A2025-09-1565 Equity Partners,Vitruvian Partners,August Global PartnersSingapore6DYNA$120MSeries A2025-09-15Charles River Ventures,First Round Capital,RoboStrategy,Amazon Industrial Innovation Fund,LG Technology VenturesUnited States7Auterion$105MSeries A2025-09-23Bessemer Venture Partners,Costanoa Ventures,Lakestar,Mosaic VenturesUnited States8CuspAI$100MSeries A2025-09-10New Enterprise Associates,Temasek,Hyundai Motor Company,Prosus Ventures,Samsung VenturesUnited Kingdom8Galaxea AI$100MSeries A2025-07-09DragonBall Capital,Meituan,Baidu Ventures,Capital Today,Cathay CapitalChina8Invisible$100MSeries A2025-09-16Vanara Capital,Backed VC,Greycroft,Acrew Capital,B&Y Venture PartnersUnited States 2025 CB Insights|29Global:Top Series B deals in Q325State of AI|Global Trends|Investment TrendsCompanyRound AmountRoundDateSelect InvestorsCountry1Nscale$1.1BSeries B2025-09-17Aker,Sandton Capital Partners,Blue Owl Capital,Dell Technologies CapitalUnited Kingdom2MiniMax$300MSeries B2025-07-14Shanghai STVC GroupChina3Distyl AI$175MSeries B2025-09-22Khosla Ventures,Lightspeed Venture Partners,DST Global,Coatue,Dell Technologies CapitalUnited States4Xelix$160MSeries B2025-07-21Insight Partners,Passion Capital,Phoenix CourtUnited Kingdom5Aura Intelligence$150MSeries B2025-09-29Undisclosed InvestorsUnited States6SiEngine Technology$139MSeries B2025-08-19Co-Stone Venture Capital,TEDA Venture Capital,Hubei Provincial High Technology Industry Investment,Jinmao CapitalChina7Armada$131MSeries B2025-07-248090 Industries,Felicis,Founders Fund,Lux Capital,M12United States8Reka AI$110MSeries B2025-07-22NVIDIA,SnowflakeUnited States9Eve$103MSeries B2025-09-30Spark Capital,Andreessen Horowitz,Lightspeed Venture Partners,Menlo VenturesUnited States10Decart$100MSeries B2025-08-07Benchmark,Sequoia Capital,Zeev Ventures,AlephUnited States10Harmonic$100MSeries B2025-07-10Kleiner Perkins,Index Ventures,Sequoia Capital,Paradigm,Ribbit CapitalUnited States10LangChain$100MSeries B2025-07-10Institutional Venture PartnersUnited States 2025 CB Insights|30Global:Top Series C deals in Q325State of AI|Global Trends|Investment TrendsCompanyRound AmountRoundDateSelect InvestorsCountry1Mistral AI$1.5BSeries C2025-09-08ASML,Andreessen Horowitz,DST Global,General Catalyst,Index VenturesFrance2Figure$1.0BSeries C2025-09-16Parkway VC,Align Ventures,Intel Capital,Brookfield Asset Management,LG Technology VenturesUnited States3Cognition$400MSeries C2025-08-14Founders Fund,Bain Capital Ventures,8VC,Elad Gil,D1 Capital PartnersUnited States4Beta Technologies$300MSeries C2025-08-26GE AerospaceUnited States5Modular AI$250MSeries C2025-09-24US Innovative Technology Fund,General Catalyst,Greylock Partners,Google Ventures,DFJ Growth FundUnited States5Rebellions$250MSeries C2025-09-30Korea Development Bank,Korelya Capital,Samsung Ventures,Arm,Lion X VenturesSouth Korea5Replit$250MSeries C2025-07-30Prysm Capital,Andreessen Horowitz,Coatue,Craft Ventures,Paul GrahamUnited States8Ambience$243MSeries C2025-07-29Andreessen Horowitz,Oak HC/FT,Kleiner Perkins,OpenAI Startup Fund,Optum VenturesUnited States9Lightelligence$210MSeries C2025-09-04CAS Star,China Mobile,China Reform Fund,Pudong VCChina10Anaconda$150MSeries C2025-07-31Insight Partners,Mubadala CapitalUnited States 2025 CB Insights|31Global:Top Series D deals in Q325State of AI|Global Trends|Investment TrendsCompanyRound AmountRoundDateRound Valuation Select InvestorsCountry1Sierra$350MSeries D2025-09-04$10.0B GreenoaksUnited States2Strive Health$300MSeries D2025-09-09N/ANew Enterprise Associates,Redpoint Ventures,CVS Health Ventures,CapitalG,Echo Health VenturesUnited States3OpenEvidence$210MSeries D2025-07-15$3.5BGoogle Ventures,Kleiner Perkins,Sequoia Capital,Coatue,Conviction CapitalUnited States4Perplexity$200MSeries D2025-09-10$20.0B Undisclosed InvestorsUnited States5AppZen$180MSeries D2025-09-22N/A Riverwood CapitalUnited States6Enveda$150MSeries D2025-09-04$1.0BPremji Invest,Baillie Gifford,Dimension Capital Management,FPV Ventures,Henry KravisUnited States6Baseten$150MSeries D2025-09-05$2.2BBond,Greylock Partners,Institutional Venture Partners,Spark Capital,01 AdvisorsUnited States8Blue J$122MSeries D2025-07-25$300MOak HC/FT,Sapphire Ventures,Ten Coves Capital,CPA.com,Intrepid Growth PartnersCanada9Aidoc$110MSeries D2025-07-23N/AGeneral Catalyst,Square Peg Capital,Hartford HealthCare,Mercy,NVenturesUnited States10Perplexity$100MSeries D2025-07-18$18.0BInstitutional Venture Partners,New Enterprise Associates,Bossa Invest,NVenturesUnited States10Cohere$100MSeries D2025-09-24$7.0B Nexxus Capital,Business Development Bank of CanadaCanada 2025 CB Insights|32Global:Top Series E deals in Q325State of AI|Global Trends|Investment TrendsCompanyRound AmountRoundDateRound Valuation Select InvestorsCountry1Anthropic$13.0BSeries F2025-09-02$183.0BFidelity Investments,ICONIQ Capital,Lightspeed Venture Partners,General Catalyst,Insight PartnersUnited States2Databricks$1.0BSeries K2025-08-19$100.0BAndreessen Horowitz,Insight Partners,MGX,Thrive Capital,WCM Investment ManagementUnited States3Groq$750MSeries E2025-09-17$6.9B Disruptive,BlackRock,Cisco,D1 Capital Partners,Neuberger BermanUnited States4Ramp$500MSeries E2025-07-30$22.5BICONIQ Growth,General Catalyst,Lightspeed Venture Partners,Sutter Hill Ventures,137 VenturesUnited States5Vercel$300MSeries F2025-09-30$9.3B Accel,GIC Group,General Catalyst,Google Ventures,Notable CapitalUnited States6Divergent$250MSeries E2025-09-15$2.3B Rochefort ManagementUnited States6EliseAI$250MSeries E2025-08-20$2.2BAndreessen Horowitz,Bessemer Venture Partners,Sapphire Ventures,Navitas CapitalUnited States8Motive$150MSeries G2025-07-30N/A Kleiner Perkins,AllianceBernsteinUnited States9Z.ai$140MSeries E2025-07-02N/A Pudong VC,Zhangjiang InnoparkChina10Nuro$97MSeries E2025-08-21$6.0B Baillie Gifford,Icehouse Ventures,Kindred Ventures,NVIDIA,Pledge VenturesUnited States 2025 CB Insights|33Headcount Trends34NEW!Fastest-growing markets by YoY headcount change 2025 CB Insights|35State of AI|Global Trends|Headcount TrendsMarketYoY headcount growth1Code documentation51.6%2Multi-agent systems&orchestration30.0=efense&national security AI copilots27.0JI agent browser infrastructure26.1ZI agent observability,evaluation,&governance24.4%6Text-to-video platforms24.3%7Multimodal AI developers23.8%8Generative AI graphic design23.1%9Generative AI music generation23.0Machine learning training data curation22.5cfisux0%Share of M&A exits by number of employeesState of AI|Global Trends|Headcount Trends62pdP,63#Q-100,14%6%6%6%81-150,81-200,3%8%7 0 ,11 212022202320242025 2025 CB Insights|36Equity deal share by number of employeesState of AI|Global Trends|Headcount Trends80 xwrqP,65%9%9Q-100,16%5%5%61-150,51-200,4%5%8%9 0 ,10%Q1 2024Q2 2024Q3 2024Q4 2024Q1 2025Q2 2025Q3 2025 2025 CB Insights|37Top deals in Q325 by valuation per employeeState of AI|Global Trends|Headcount TrendsCompanyValuation per EmployeeDeal ValuationHeadcountDeal Date1Figure$104.3M$39.0B3742025-09-162Cognition$98.1M$10.2B1042025-08-143Anthropic$82.1M$183.0B22302025-09-024OpenEvidence$72.9M$3.5B482025-07-155Decart$43.7M$3.1B712025-08-076Sierra$32.5M$10.0B3082025-09-047Mistral AI$29.5M$13.2B4492025-09-088Harmonic$24.3M$875M362025-07-109Irregular$22.5M$450M202025-09-1710Baseten$19.4M$2.1B1112025-09-05 2025 CB Insights|38Distribution of deals by stage and headcountState of AI|Global Trends|Headcount Trends50employees51-100employees101-150employees151-200employees200 employeesEarly-stage372deals439710Mid-stage3145171621Late-stage20196432Other273325 2025 CB Insights|39Share of deals by Commercial Maturity and headcountState of AI|Global Trends|Headcount TrendsCommercial Maturity is a CB Insights proprietary metric.It measures a private companys current ability to compete for customers or serve as a partner.Ranging from Emerging to Established,it provides a clear,quantitative measurement for where a company lies in its development and growth process.2025 CB Insights|4025%8WqVB)8P51-100101-150151-200200 EmergingInitial research&developmentValidatingTesting and refining productDeployingGrowing commercial distributionScalingExpanding to additional marketsEstablishedMajor market presenceemployeesemployeesemployeesemployeesemployeesUnicorns41Quarterly new&total unicorns(private companies valued at$1b )State of AI|Global Trends|Unicorns14121618New Unicorns,18282289301316Total Unicorns,331Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|42State of AI|Global Trends|UnicornsUS23116 NewAsia472 NewEurope430 NewCanada50 NewOceania20 NewLatAm30 NewNew&total unicorns by global region in Q325 2025 CB Insights|43Quarterly new unicorns by global regionState of AI|Global Trends|Unicorns11111211US,16112Asia,22135Europe,0LatAm,0Canada,0Oceania,0Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|44Global:Top unicorn births in Q325State of AI|Global Trends|UnicornsCompanyLatest Valuation Country1Decart$3.1BUnited States2Baseten$2.2BUnited States3FieldAI$2.0BUnited States4Distyl AI$1.8BUnited States4Lovable$1.8BUnited States6Modular AI$1.6BUnited States7Anaconda$1.5BUnited States7Fal$1.5BUnited States7Y$1.5BUnited States10Lila Sciences$1.2BUnited States 2025 CB Insights|45Global:Top unicorns by valuation in Q325State of AI|Global Trends|UnicornsCompanyLatest Valuation Country1ByteDance$300.0BChina1OpenAI$300.0BUnited States3Anthropic$183.0BUnited States4Stripe$106.7BUnited States5Databricks$100.0BUnited States6xAI$75.0BUnited States7Canva$42.0BAustralia8Figure$39.0BUnited States9Safe Superintelligence$32.0BUnited States10Anduril$30.5BUnited States 2025 CB Insights|46Exit Trends47Quarterly exitsState of AI|Global Trends|Exit Trends61638IPO,1383133108181M&A,1722112SPAC,1Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|48Percent of quarterly exits by global regionState of AI|Global Trends|Exit Trends54dRQFDB9A7BB9GPEUW%US,59!$#%9%9%7%Asia,8%$%(42B53&A7221&%Europe,28%LatAm,1%6%9%6%9%5nada,3rica,0%Oceania,1%All Other Regions,0%Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|49Global:Top M&A exits in Q325State of AI|Global Trends|Exit TrendsCompanyRound ValuationAcquirerCountry1Sana Labs$1.1B WorkdaySweden1Statsig$1.1B OpenAIUnited States3Cognigy$955M NICEGermany4NetBrain Technologies$750M BlackstoneUnited States5The Browser Company$610M AtlassianUnited States6Lakera$300M Check PointUnited States6Prompt Security$300M SentinelOneUnited States6VideoVerse$300M Minute MediaUnited States9HKB Mortgage$229M HulicJapan10Daedalean$223M DestinusSwitzerland 2025 CB Insights|50Global:Top IPOs in Q325State of AI|Global Trends|Exit TrendsCompanyRound Valuation Country1Accelerant$6.4BUnited States2Figure$5.3BUnited States3Pattern$2.5BUnited States4HeartFlow$1.5BUnited States5WhiteFiber$619MUnited States6Ambiq$423MUnited States7Carlsmed$398MUnited States8Robot Consulting$184MJapan9KNOREX$122MUnited States10The GrowHub$101MSingapore 2025 CB Insights|51Global:Top SPACs in Q325State of AI|Global Trends|Exit TrendsCompanyRound Valuation Select InvestorsCountry1Kodiak Robotics$2.5BAres Acquisition Corporation IIUnited States 2025 CB Insights|52Investors53Global:Top investors by company count in Q325State of AI|Global Trends|InvestorsInvestorCompany Count Investor GroupCountry1Pioneer Fund38VCUnited States2Andreessen Horowitz24VCUnited States3General Catalyst23VCUnited States4Khosla Ventures16VCUnited States4Lightspeed Venture Partners16VCUnited States6Antler14VCSingapore6Insight Partners14Private EquityUnited States8Bessemer Venture Partners13VCUnited States9New Enterprise Associates12VCUnited States10NVentures11CVCUnited States 2025 CB Insights|54Global:Top VCs by company count in Q325State of AI|Global Trends|InvestorsInvestorCompany Count Country1Pioneer Fund38United States2Andreessen Horowitz24United States3General Catalyst23United States4Khosla Ventures16United States4Lightspeed Venture Partners16United States6Antler14Singapore7Bessemer Venture Partners13United States8New Enterprise Associates12United States9Sequoia Capital10United States9Accel10United States 2025 CB Insights|55Global:Top corporate VCs(CVCs)by company count in Q325State of AI|Global Trends|InvestorsInvestorCompany Count Country1NVentures11United States2Salesforce Ventures10United States3Google Ventures9United States4Prosus Ventures6Netherlands5Nissay Capital5Japan6In-Q-Tel4United States6Intel Capital4United States6SBI Investment4Japan6Samsung NEXT4United States6Smilegate Investment4South Korea 2025 CB Insights|56Geographic Trends57US Trends58Quarterly funding&dealsState of AI|Geographic Trends|US Trends$17.7B$16.6B$17.3B$17.4B$16.0B$11.9B$8.5B$7.8B$23.3B$9.2B$9.9B$7.7B$11.3B$18.0B$12.6B$39.4B$55.8B$40.6BFunding$38.8B608542671622755607566488728639665569738686733676819906Deals65202004006008001,0001,200$0.0B$10.0B$20.0B$30.0B$40.0B$50.0B$60.0BQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|59Annual percent of deals by deal stageState of AI|Geographic Trends|US Trends64iwurly-stage,73%Mid-stage,12%7%5%Late-stage,7%8%8%8%Other,8 212022202320242025YTD 2025 CB Insights|60US:Top equity deals in Q325State of AI|Geographic Trends|US TrendsCompanyRound AmountRoundDateRound Valuation Select Investors1Anthropic$13.0BSeries F2025-09-02$183.0BFidelity Investments,ICONIQ Capital,Lightspeed Venture Partners,General Catalyst,Insight Partners2OpenAI$8.3BPrivate Equity2025-08-01$300.0BDragoneer Investment Group,Andreessen Horowitz,Sequoia Capital,Altimeter Capital,Bossa Invest3Databricks$1.0BSeries K2025-08-19$100.0BAndreessen Horowitz,Insight Partners,MGX,Thrive Capital,WCM Investment Management3Figure$1.0BSeries C2025-09-16$39.0BParkway VC,Align Ventures,Intel Capital,Brookfield Asset Management,LG Technology Ventures5Groq$750MSeries E2025-09-17$6.9B Disruptive,BlackRock,Cisco,D1 Capital Partners,Neuberger Berman6Ramp$500MSeries E2025-07-30$22.5BICONIQ Growth,General Catalyst,Lightspeed Venture Partners,Sutter Hill Ventures,137 Ventures7Cognition$400MSeries C2025-08-14$10.2BFounders Fund,Bain Capital Ventures,8VC,Elad Gil,D1 Capital Partners8Sierra$350MSeries D2025-09-04$10.0B Greenoaks9FieldAI$314MSeries A2025-08-20$2.0B Bezos Expeditions,Prysm Capital,Temasek,BHP Ventures 2025 CB Insights|61Quarterly exitsState of AI|Geographic Trends|US Trends1314IPO,8436662105M&A,102111SPAC,1Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|62US:Top investors by company count in Q325State of AI|Geographic Trends|US TrendsInvestorCompany Count Investor Group1Pioneer Fund38VC2Andreessen Horowitz24VC3General Catalyst23VC4Khosla Ventures16VC4Lightspeed Venture Partners16VC6Insight Partners14Private Equity7Bessemer Venture Partners13VC8New Enterprise Associates12VC9NVentures11CVC10Sequoia Capital10VC 2025 CB Insights|63Silicon Valley64Quarterly funding&dealsState of AI|Geographic Trends|Silicon Valley$10.4B$7.9B$8.4B$8.6B$8.0B$5.3B$3.1B$2.3B$19.9B$5.2B$4.9B$2.6B$8.5B$11.2B$6.5B$33.0B$46.1B$29.7BFunding$30.1B230193228227264227195149281213227187312258259249335418Deals2720100200300400500$0.0B$5.0B$10.0B$15.0B$20.0B$25.0B$30.0B$35.0B$40.0B$45.0BQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|65Annual percent of deals by deal stageState of AI|Geographic Trends|Silicon Valley62fyurly-stage,78#%Mid-stage,11%8%Late-stage,7%7%5%Other,4 212022202320242025YTD 2025 CB Insights|66Silicon Valley:Top equity deals in Q325State of AI|Geographic Trends|Silicon ValleyCompanyRound AmountRoundDateRound Valuation Select Investors1Anthropic$13.0BSeries F2025-09-02$183.0BFidelity Investments,ICONIQ Capital,Lightspeed Venture Partners,General Catalyst,Insight Partners2OpenAI$8.3BPrivate Equity2025-08-01$300.0BDragoneer Investment Group,Andreessen Horowitz,Sequoia Capital,Altimeter Capital,Bossa Invest3Databricks$1.0BSeries K2025-08-19$100.0BAndreessen Horowitz,Insight Partners,MGX,Thrive Capital,WCM Investment Management3Figure$1.0BSeries C2025-09-16$39.0BParkway VC,Align Ventures,Intel Capital,Brookfield Asset Management,LG Technology Ventures5Groq$750MSeries E2025-09-17$6.9B Disruptive,BlackRock,Cisco,D1 Capital Partners,Neuberger Berman 2025 CB Insights|67New York68Quarterly funding&dealsState of AI|Geographic Trends|New York$3.1B$1.6B$2.7B$2.2B$1.1B$1.1B$0.7B$0.9B$0.7B$1.6B$1.2B$1.8B$0.7B$4.0B$1.2B$1.0B$1.7B$2.3BFunding$3.2B91741017810167767110681907811011512072110119Deals104020406080100120140160$0.0B$0.5B$1.0B$1.5B$2.0B$2.5B$3.0B$3.5B$4.0B$4.5BQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|69Annual percent of deals by deal stageState of AI|Geographic Trends|New York65xywrly-stage,70%9%8%Mid-stage,14%8%6%6%Late-stage,7%8%8%7%6%Other,9 212022202320242025YTD 2025 CB Insights|70New York:Top equity deals in Q325State of AI|Geographic Trends|New YorkCompanyRound AmountRoundDateRound Valuation Select Investors1Ramp$500MSeries E2025-07-30$22.5BICONIQ Growth,General Catalyst,Lightspeed Venture Partners,Sutter Hill Ventures,137 Ventures2Cognition$400MSeries C2025-08-14$10.2BFounders Fund,Bain Capital Ventures,8VC,Elad Gil,D1 Capital Partners3Emergence AI$300MCorporate Minority2025-09-24N/A Russell AI Labs4EliseAI$250MSeries E2025-08-20$2.2BAndreessen Horowitz,Bessemer Venture Partners,Sapphire Ventures,Navitas Capital5Aidoc$110MSeries D2025-07-23N/AGeneral Catalyst,Square Peg Capital,Hartford HealthCare,Mercy,NVentures 2025 CB Insights|71Los Angeles72Quarterly funding&dealsState of AI|Geographic Trends|Los Angeles$0.2B$1.8B$0.4B$0.4B$0.7B$0.5B$1.0B$1.9B$0.2B$0.3B$0.2B$0.4B$0.1B$0.5B$1.8B$0.7B$0.9B$3.1BFunding$1.1B303343393931382536344334283432403437Deals25051015202530354045$0.0B$0.5B$1.0B$1.5B$2.0B$2.5B$3.0BQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|73Annual percent of deals by deal stageState of AI|Geographic Trends|Los Angeles72psirly-stage,59%9%Mid-stage,15%8%8%5%Late-stage,11%6%Other,15 212022202320242025YTD 2025 CB Insights|74Los Angeles:Top equity deals in Q325State of AI|Geographic Trends|Los AngelesCompanyRound AmountRoundDateRound Valuation Select Investors1FieldAI$314MSeries A2025-08-20$2.0B Bezos Expeditions,Prysm Capital,Temasek,BHP Ventures2Vercel$300MSeries F2025-09-30$9.3B Accel,GIC Group,General Catalyst,Google Ventures,Notable Capital3Divergent$250MSeries E2025-09-15$2.3B Rochefort Management4MarqVision$48MSeries B2025-09-15N/APeak XV Partners,Altos Ventures,Atinum Investment,Smilegate Investment,Y Combinator5ProRata.ai$40MSeries B2025-09-05N/A Touring Capital,Bold Capital,MVP Ventures,Mayfield,Revolution Ventures 2025 CB Insights|75Boston76Quarterly funding&dealsState of AI|Geographic Trends|Boston$1.2B$1.2B$1.5B$1.5B$1.3B$1.3B$0.7B$0.7B$0.3B$0.4B$1.0B$0.6B$0.4B$0.4B$0.4B$0.8B$1.1B$0.6BFunding$0.6B385147474950383739404938373837514140Deals27010203040506070$0.0B$0.2B$0.4B$0.6B$0.8B$1.0B$1.2B$1.4BQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|77Annual percent of deals by deal stageState of AI|Geographic Trends|Boston64dpprly-stage,53 %Mid-stage,25%7%7%6%Late-stage,6%5%8%Other,17 212022202320242025YTD 2025 CB Insights|78Boston:Top equity deals in Q325State of AI|Geographic Trends|BostonCompanyRound AmountRoundDateRound Valuation Select Investors1Lila Sciences$235MSeries A2025-09-15$1.2BBraidwell,Collective Global,General Catalyst,Ark Ventures,Flagship Pioneering2OpenEvidence$210MSeries D2025-07-15$3.5BGoogle Ventures,Kleiner Perkins,Sequoia Capital,Coatue,Conviction Capital3Blue Water Autonomy$50MSeries A2025-08-26N/A Google Ventures,Eclipse,Impatient Ventures,Riot Ventures4Ketryx$39MSeries B2025-09-04N/ATransformation Capital,Lightspeed Venture Partners,E14 Fund,Ubiquity Ventures,53 Stations5Method AI$20MSeries A2025-08-21N/A Cleveland Clinic,JobsOhio 2025 CB Insights|79Seattle80Quarterly funding&dealsState of AI|Geographic Trends|Seattle$397M$296M$343M$360M$484M$253M$228M$107M$80M$224M$278M$279M$56M$125M$489M$298M$627M$965MFunding$184M201229232317251420242130242431213033Deals200510152025303540$0.0B$0.2B$0.4B$0.6B$0.8B$1.0BQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|81Annual percent of deals by deal stageState of AI|Geographic Trends|Seattle74sxyrly-stage,71%9%Mid-stage,18%6%5%Late-stage,4%7%8%8%8%Other,7 212022202320242025YTD 2025 CB Insights|82Seattle:Top equity deals in Q325State of AI|Geographic Trends|SeattleCompanyRound AmountRoundDateRound Valuation Select Investors1Augmodo$38MSeries A2025-07-10N/ATQ Ventures,Lerer Hippeau,New Fare Partners,Arena Holdings,Interlace Ventures2Dropzone AI$37MSeries B2025-07-28N/ATheory Ventures,Decibel Partners,In-Q-Tel,Madrona Venture Group,Pioneer Square Labs3Vouched$17MSeries A2025-09-04N/A SpringRock Ventures,BHG VC4Envive$15MSeries A2025-09-16N/A FUSE4Everlyn$15MSeries A2025-08-31$250M Aethir,MH Ventures,Nesa,Selini Capital,Baseline 2025 CB Insights|83Austin84Quarterly funding&dealsState of AI|Geographic Trends|Austin$413M$263M$382M$436M$347M$202M$112M$188M$107M$109M$200M$221M$38M$112M$313M$206M$1,371M$140MFunding$453M201514182316141617171721121327241715Deals16051015202530$0.0B$0.2B$0.4B$0.6B$0.8B$1.0B$1.2B$1.4BQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|85Annual percent of deals by deal stageState of AI|Geographic Trends|Austin67ryyrly-stage,58%8%Mid-stage,15%9%6%6%Late-stage,15%Other,13 212022202320242025YTD 2025 CB Insights|86Austin:Top equity deals in Q325State of AI|Geographic Trends|AustinCompanyRound AmountRoundDateRound Valuation Select Investors1Anaconda$150MSeries C2025-07-31$1.5B Insight Partners,Mubadala Capital2Augment$85MSeries A2025-09-04N/A Redpoint Ventures,8VC,Autotech Ventures,Shopify Ventures3SEON$80MSeries C2025-09-16N/ASixth Street Growth,Institutional Venture Partners,Creandum,Firebolt Ventures,Hearst4Pattern Bioscience$44MSeries D2025-08-18N/A AMR Action Fund,Apalachee Ventures,Illumina Ventures5Hello Patient$23MSeries A2025-09-04N/AScale Venture Partners,8VC,Bling Capital,Max Ventures,FirstLook Partners 2025 CB Insights|87Miami88Quarterly funding&dealsState of AI|Geographic Trends|Miami$18M$209M$193M$569M$64M$110M$83M$61M$63M$23M$99M$49M$53M$48M$52M$122M$164M$139MFunding$74M12815141920131022131410161213151612Deals16051015202530$0M$100M$200M$300M$400M$500MQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|89Annual percent of deals by deal stageState of AI|Geographic Trends|Miami73wrly-stage,64%6%5%7%Mid-stage,11%Late-stage,5%7%Other,20 212022202320242025YTD 2025 CB Insights|90Miami:Top equity deals in Q325State of AI|Geographic Trends|MiamiCompanyRound AmountRoundDateRound Valuation Select Investors1Imagene AI$23MSeries B2025-07-01N/A Larry Ellison,Aguras Pathology Investments2Osigu$10MSeries B2025-08-28N/A Eos Venture Partners2Togal.AI$10MConvertible Note2025-08-27N/A Undisclosed Investors4EchoTwin AI$8MSeed VC2025-09-24N/AMetis Ventures,Automotive Ventures,Eksim Ventures,HL Ventures,Supernova5Kira$7MSeed VC2025-08-25N/ABlockchange Ventures,Credibly Neutral,Grit Ventures,Stellar,VamosVentures 2025 CB Insights|91Philadelphia92Quarterly funding&dealsState of AI|Geographic Trends|Philadelphia$53M$360M$95M$219M$298M$245M$113M$157M$146M$202M$248M$211M$294M$182M$178M$166M$508M$805MFunding$848M231216203222162352353331402841283846Deals35010203040506070$0M$100M$200M$300M$400M$500M$600M$700M$800M$900MQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|93Annual percent of deals by deal stageState of AI|Geographic Trends|Philadelphia82rly-stage,84%8%9%7%Mid-stage,5%Late-stage,8%8%6%Other,3 212022202320242025YTD 2025 CB Insights|94Philadelphia:Top equity deals in Q325State of AI|Geographic Trends|PhiladelphiaCompanyRound AmountRoundDateRound Valuation Select Investors1Quavo$300MGrowth Equity2025-07-22N/A Spectrum Equity2Lovable$200MSeries A2025-07-17$1.8B Accel,Creandum,Visionaries Club,byFounders,Hummingbird3Decart$100MSeries B2025-08-07$3.1B Benchmark,Sequoia Capital,Zeev Ventures,Aleph4Maisa AI$25MSeed VC2025-08-28N/ACreandum,NFX,Village Global,Forgepoint Capital International,Hoxton Ventures4Maisa AI$25MConvertible Note2025-07-24N/A Undisclosed Investors 2025 CB Insights|95Canada Trends96Quarterly funding&dealsState of AI|Geographic Trends|Canada Trends$122M$1,049M$572M$181M$909M$593M$159M$238M$82M$744M$306M$247M$226M$723M$178M$937M$375M$753MFunding$439M413639425342273637444140273734312133Deals21010203040506070$0.0B$0.2B$0.4B$0.6B$0.8B$1.0B$1.2BQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|97Annual percent of deals by deal stageState of AI|Geographic Trends|Canada Trends77wxrly-stage,75%7%Mid-stage,13%6%7%Late-stage,12%7%8%7%Other,0 212022202320242025YTD 2025 CB Insights|98Canada:Top equity deals in Q325State of AI|Geographic Trends|Canada TrendsCompanyRound AmountRoundDateRound Valuation Select Investors1Blue J$122MSeries D2025-07-25$300MOak HC/FT,Sapphire Ventures,Ten Coves Capital,CPA.com,Intrepid Growth Partners2Cohere$100MSeries D2025-09-24$7.0B Nexxus Capital,Business Development Bank of Canada3BinSentry$50MSeries C2025-08-11N/A Lead Edge Capital3Valence$50MSeries B2025-09-25N/A Bessemer Venture Partners5Congruence Therapeutics$32MSeries A2025-09-04N/AAmplitude Ventures,Alexandria Venture Investments,BDC Capital,Driehaus Capital Management,Investissement Quebec64AG Robotics$29MSeries B2025-07-16N/A Astanor,Cibus Fund,BDC Capital,Emmertech,inBC7BinSentry$20MSeries B2025-07-16N/A BDC Venture Capital8Astrus$8MConvertible Note2025-07-01N/AKhosla Ventures,1517 Fund,Alumni Ventures,MVP Ventures,Drive Capital9Hyper$6MSeed VC2025-07-21N/AEniac Ventures,Alumni Ventures,Blue Moon,Four Acres Capital,GreatPoint Ventures9TATO$6MSeed VC2025-09-22N/A Ridge Ventures 2025 CB Insights|99Quarterly exitsState of AI|Geographic Trends|Canada Trends111IPO,01418M&A,61SPAC,0Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|100Canada:Top investors by company count in Q325State of AI|Geographic Trends|Canada TrendsInvestorCompany Count Investor Group1Radical Ventures4VC2Georgian3VC2Portage Ventures3VC4BDC Capital2VC4Brookfield Asset Management2Asset/investment management4Humanoid Global Holdings2VC4Kaz Nejatian2Angel4Mistral Venture Partners2VC4Ontario Teachers Pension Plan2Asset/investment management4Ripple Ventures2VC4Shopify Ventures2CVC4Sixty Degree Capital2VC 2025 CB Insights|101Asia Trends102Quarterly funding&dealsState of AI|Geographic Trends|Asia Trends$5.1B$5.4B$5.5B$5.2B$4.3B$3.3B$2.6B$1.7B$1.1B$2.0B$2.2B$3.0B$3.1B$3.2B$2.3B$3.0B$2.1B$2.2BFunding$2.9B362347464406428359379276256287300309279292379339317337Deals2970100200300400500600$0.0B$1.0B$2.0B$3.0B$4.0B$5.0BQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|103Annual percent of deals by deal stageState of AI|Geographic Trends|Asia Trends68hqsrly-stage,71!%Mid-stage,16%8%8%8%8%Late-stage,10%5%Other,3 212022202320242025YTD 2025 CB Insights|104Asia:Top equity deals in Q325State of AI|Geographic Trends|Asia TrendsCompanyRound AmountRoundDateRound Valuation Select InvestorsCountry1MiniMax$300MSeries B2025-07-14$3.7B Shanghai STVC GroupChina2Rebellions$250MSeries C2025-09-30$1.4BKorea Development Bank,Korelya Capital,Samsung Ventures,Arm,Lion X VenturesSouth Korea3Lightelligence$210MSeries C2025-09-04N/A CAS Star,China Mobile,China Reform Fund,Pudong VCChina4UltraGreen$188MSeries A2025-09-15$1.3B 65 Equity Partners,Vitruvian Partners,August Global Partners Singapore5Z.ai$140MSeries E2025-07-02N/A Pudong VC,Zhangjiang InnoparkChina6SiEngine Technology$139MSeries B2025-08-19N/ACo-Stone Venture Capital,TEDA Venture Capital,Hubei Provincial High Technology Industry Investment,Jinmao CapitalChina7TARS$125MSeed VC2025-07-08$822MMeituan,Linear Venture,Xiang He Capital,C&D Emerging Investment,CIVCChina8Galaxea AI$100MSeries A2025-07-09N/ADragonBall Capital,Meituan,Baidu Ventures,Capital Today,Cathay CapitalChina9Lisuan Tech$70MCorporate Minority2025-08-31$589M Dosilicon,Hengtong Group,Daohe Long-term InvestmentChina9ZERON$70MSeries A2025-07-23N/AEstar Capital,MinxiXinghang State-owned Investment&Operation,Momenta,Skyview Fund,Caitong CapitalChina 2025 CB Insights|105Quarterly exitsState of AI|Geographic Trends|Asia Trends41012IPO,5612911M&A,111SPAC,0Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|106Asia:Top investors by company count in Q325State of AI|Geographic Trends|Asia TrendsInvestorCompany Count Investor GroupCountry1Antler14VCSingapore2HongShan8VCChina3Peak XV Partners6VCIndia4Headline5VCJapan4InnoAngel Fund5VCChina4Nissay Capital5CVCJapan 2025 CB Insights|107China108Quarterly funding&dealsState of AI|Geographic Trends|China$4.4B$3.9B$3.7B$3.8B$1.9B$1.9B$1.7B$1.2B$0.5B$1.3B$1.3B$1.6B$2.3B$0.7B$1.4B$1.9B$0.8B$1.3BFunding$1.6B1821902441771651321241068799104112789510610196137Deals112050100150200250$0.0B$1.0B$2.0B$3.0B$4.0B$5.0BQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|109Annual percent of deals by deal stageState of AI|Geographic Trends|China61Tcerly-stage,63)6#%Mid-stage,24%8%9%8%8%Late-stage,12%Other,2 212022202320242025YTD 2025 CB Insights|110China:Top equity deals in Q325State of AI|Geographic Trends|ChinaCompanyRound AmountRoundDateRound Valuation Select Investors1MiniMax$300MSeries B2025-07-14$3.7B Shanghai STVC Group2Lightelligence$210MSeries C2025-09-04N/A CAS Star,China Mobile,China Reform Fund,Pudong VC3Z.ai$140MSeries E2025-07-02N/A Pudong VC,Zhangjiang Innopark4SiEngine Technology$139MSeries B2025-08-19N/ACo-Stone Venture Capital,TEDA Venture Capital,Hubei Provincial High Technology Industry Investment,Jinmao Capital5TARS$125MSeed VC2025-07-08$822MMeituan,Linear Venture,Xiang He Capital,C&D Emerging Investment,CIVC 2025 CB Insights|111India112Quarterly funding&dealsState of AI|Geographic Trends|India$69M$162M$398M$454M$1,114M$290M$116M$131M$108M$174M$361M$322M$198M$230M$305M$186M$382M$358MFunding$229M403656568145474036362923284955424237Deals410102030405060708090100$0M$200M$400M$600M$800M$1,000M$1,200MQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|113Annual percent of deals by deal stageState of AI|Geographic Trends|India83rly-stage,81%9%8%Mid-stage,9%8%6%Late-stage,7%Other,3 212022202320242025YTD 2025 CB Insights|114India:Top equity deals in Q325State of AI|Geographic Trends|IndiaCompanyRound AmountRoundDateRound Valuation Select Investors1Amnex$53MUndisclosed2025-08-18N/A Wealth Company2Darwinbox$40MSeries E2025-08-14N/A Ontario Teachers Pension Plan3QpiAI$32MSeries A2025-07-16$162MAvataar Venture Partners,Department of Science&Technology(India)4Ripplr$23MSeries C2025-09-13N/A Sojitz,3one4 Capital Partners,Trifecta Capital Advisors5Netrasemi$12MSeries A2025-07-24N/A Unicorn India Ventures,Zoho,Maithan Alloys 2025 CB Insights|115Singapore116Quarterly funding&dealsState of AI|Geographic Trends|Singapore$15M$219M$551M$356M$342M$358M$68M$86M$53M$118M$57M$70M$173M$203M$113M$59M$142M$87MFunding$354M202334243225181723212024242330252519Deals2905101520253035$0M$100M$200M$300M$400M$500M$600MQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|117Annual percent of deals by deal stageState of AI|Geographic Trends|Singapore80rly-stage,79%6%Mid-stage,8%Late-stage,4%9%6%Other,8 212022202320242025YTD 2025 CB Insights|118Singapore:Top equity deals in Q325State of AI|Geographic Trends|SingaporeCompanyRound AmountRoundDateRound Valuation Select Investors1UltraGreen$188MSeries A2025-09-15$1.3B 65 Equity Partners,Vitruvian Partners,August Global Partners2Haivivi$28MSeries A2025-08-25N/ACICC Capital,HongShan,Huashan Capital,JOY Capital,Brizan Investments3AND Global$21MSeries B2025-08-18N/AAEON Financial Service,International Finance Corporation,Marubeni,SBI Holdings,Premium Group4Kite AI$18MSeries A2025-09-02N/A General Catalyst,PayPal Ventures,8VC,Alchemy,Alumni Ventures5Atomionics$13MSeed VC2025-09-08N/A Paspalis,SGInnovate,Wavemaker Partners,BHP Ventures,In-Q-Tel 2025 CB Insights|119Japan120Quarterly funding&dealsState of AI|Geographic Trends|Japan$261M$108M$384M$199M$102M$373M$221M$70M$192M$104M$269M$107M$134M$318M$126M$73M$83M$158MFunding$235M432128403752613844526051604973576064Deals490102030405060708090$0.0B$0.1B$0.1B$0.2B$0.2B$0.3B$0.3B$0.4B$0.4BQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|121Annual percent of deals by deal stageState of AI|Geographic Trends|Japan58RVYrly-stage,64!%Mid-stage,16 %Late-stage,16%7%9%Other,3 212022202320242025YTD 2025 CB Insights|122Japan:Top equity deals in Q325State of AI|Geographic Trends|JapanCompanyRound AmountRoundDateRound Valuation Select Investors1SkyDrive$58MSeries C2025-07-04N/AMUFG Bank,ITOCHU Technology Ventures,Kansai Electric Power,NHK Spring,Obayashi Corporation2LegalOn Technologies$50MSeries E2025-07-17N/AGoldman Sachs Alternatives,WiL,Mori Hamada&Matsumoto3T2$34MSeries A2025-08-18N/AEnergy&Environment Investment,Mitsubishi Estate,Mitsui Sumitomo Insurance,Mitsui-Soko Group,Usami Koyu4Helpfeel$18MSeries E2025-08-27N/AGlobal Brain,Fukoku Mutual Life Insurance Company,Japan Post Investment,SMBC Venture Capital5Japan AI$13MSeries B2025-07-23N/AJAFCO,Branding Technology,FCE Publishing,Nissay Capital,Resona Capital 2025 CB Insights|123Europe Trends124Quarterly funding&dealsState of AI|Geographic Trends|Europe Trends$2.2B$4.5B$2.8B$3.6B$4.7B$3.3B$2.5B$2.3B$1.9B$2.6B$2.2B$2.3B$2.2B$4.3B$2.7B$2.8B$3.8B$4.7BFunding$5.4B291275248299384300253296348284255277289317344353306353Deals279050100150200250300350400450500$0.0B$1.0B$2.0B$3.0B$4.0B$5.0B$6.0BQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|125Annual percent of deals by deal stageState of AI|Geographic Trends|Europe Trends77rly-stage,78%8%9%Mid-stage,11%7%6%Late-stage,6%6%Other,6 212022202320242025YTD 2025 CB Insights|126Europe:Top equity deals in Q325State of AI|Geographic Trends|Europe TrendsCompanyRound AmountRoundDateRound Valuation Select InvestorsCountry1Mistral AI$1.5BSeries C2025-09-08$13.2BASML,Andreessen Horowitz,DST Global,General Catalyst,Index VenturesFrance2Nscale$1.1BSeries B2025-09-17N/AAker,Sandton Capital Partners,Blue Owl Capital,Dell Technologies CapitalUnited Kingdom3Xelix$160MSeries B2025-07-21N/A Insight Partners,Passion Capital,Phoenix CourtUnited Kingdom4CuspAI$100MSeries A2025-09-10$520MNew Enterprise Associates,Temasek,Hyundai Motor Company,Prosus Ventures,Samsung VenturesUnited Kingdom4Noma Security$100MSeries B2025-07-31N/AEvolution Equity Partners,Ballistic Ventures,Glilot Capital PartnersIsrael6Exodigo$96MSeries B2025-07-16$700MGreenfield Partners,Zeev Ventures,10D Ventures,Jibe Ventures,Square Peg CapitalIsrael7Charm Therapeutics$80MSeries B2025-09-02N/ANew Enterprise Associates,SR One,F-Prime Capital,Khosla Ventures,NVenturesUnited Kingdom7Irregular$80MSeries A2025-09-17$450MRedpoint Ventures,Sequoia Capital,Swish Ventures,Assaf Rappaport,Ofir EhrlichIsrael9Makersite$70MSeries B2025-07-22N/ALightrock,Partech,Hitachi Ventures,Kompas,Planet A VenturesGermany10Ultromics$55MSeries C2025-07-31N/AAllegis Capital,Legal&General,Lightrock,Google Ventures,Oxford Science EnterprisesUnited Kingdom 2025 CB Insights|127Quarterly exitsState of AI|Geographic Trends|Europe Trends121IPO,027463248M&A,501SPAC,0Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|128Europe:Top investors by company count in Q325State of AI|Geographic Trends|Europe TrendsInvestorCompany Count Investor GroupCountry1Kima Ventures8AngelFrance1Seedcamp8VCUnited Kingdom3Bpifrance6Asset/investment managementFrance3Prosus Ventures6CVCNetherlands5Creandum5VCSweden6Cherry Ventures4VCGermany6JME Venture Capital4VCSpain6Purple Ventures4VCCzech Republic6The Twenty Minute VC4VCUnited Kingdom6Thomas Wolf4AngelNetherlands 2025 CB Insights|129United Kingdom130Quarterly funding&dealsState of AI|Geographic Trends|United Kingdom$0.4B$1.9B$1.1B$0.9B$1.1B$0.6B$0.6B$0.7B$0.5B$1.0B$0.5B$0.3B$0.4B$1.5B$0.7B$0.8B$1.5B$1.0BFunding$2.0B8269716711985637199718771848499918199Deals70020406080100120140160$0.0B$0.5B$1.0B$1.5B$2.0B$2.5BQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|131Annual percent of deals by deal stageState of AI|Geographic Trends|United Kingdom80rly-stage,76%7%9%Mid-stage,12%5%5%Late-stage,6%5%5%7%6%Other,6 212022202320242025YTD 2025 CB Insights|132United Kingdom:Top equity deals in Q325State of AI|Geographic Trends|United KingdomCompanyRound AmountRoundDateRound Valuation Select Investors1Nscale$1.1BSeries B2025-09-17N/AAker,Sandton Capital Partners,Blue Owl Capital,Dell Technologies Capital2Xelix$160MSeries B2025-07-21N/A Insight Partners,Passion Capital,Phoenix Court3CuspAI$100MSeries A2025-09-10$520MNew Enterprise Associates,Temasek,Hyundai Motor Company,Prosus Ventures,Samsung Ventures4Charm Therapeutics$80MSeries B2025-09-02N/ANew Enterprise Associates,SR One,F-Prime Capital,Khosla Ventures,NVentures5Ultromics$55MSeries C2025-07-31N/AAllegis Capital,Legal&General,Lightrock,Google Ventures,Oxford Science Enterprises 2025 CB Insights|133Germany134Quarterly funding&dealsState of AI|Geographic Trends|Germany$301M$353M$319M$503M$233M$500M$153M$251M$459M$255M$395M$445M$229M$599M$664M$231M$397M$1,361MFunding$246M384130433740273338262936312839453444Deals3305101520253035404550$0.0B$0.2B$0.4B$0.6B$0.8B$1.0B$1.2B$1.4B$1.6BQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|135Annual percent of deals by deal stageState of AI|Geographic Trends|Germany80rly-stage,82%8%9%Mid-stage,13%5%7%Late-stage,4%Other,2 212022202320242025YTD 2025 CB Insights|136Germany:Top equity deals in Q325State of AI|Geographic Trends|GermanyCompanyRound AmountRoundDateRound Valuation Select Investors1Makersite$70MSeries B2025-07-22N/A Lightrock,Partech,Hitachi Ventures,Kompas,Planet A Ventures2Fernride$21MSeries A2025-09-04N/A Helantic,10 x Founders,DeepTech&Climate Fonds,Hensoldt3MOTOR Ai$20MSeed VC2025-07-14N/A Segenia Capital,eCAPITAL4Kertos$17MSeries A2025-09-17N/APortage Ventures,10 x Founders,Pi Labs,Redstone,Seed Speed Ventures5Born$15MSeries A2025-09-10N/A Accel,Laton Ventures,Tencent 2025 CB Insights|137France138Quarterly funding&dealsState of AI|Geographic Trends|France$0.2B$1.1B$0.0B$0.5B$0.7B$0.4B$0.2B$0.2B$0.3B$0.6B$0.6B$0.7B$0.4B$1.0B$0.3B$0.7B$0.2B$0.4BFunding$1.7B272913244331292740362037312534452338Deals1901020304050$0.0B$0.2B$0.4B$0.6B$0.8B$1.0B$1.2B$1.4B$1.6B$1.8BQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|139Annual percent of deals by deal stageState of AI|Geographic Trends|France74xrly-stage,80%9%Mid-stage,14%Late-stage,4%5%Other,3 212022202320242025YTD 2025 CB Insights|140France:Top equity deals in Q325State of AI|Geographic Trends|FranceCompanyRound AmountRoundDateRound Valuation Select Investors1Mistral AI$1.5BSeries C2025-09-08$13.2BASML,Andreessen Horowitz,DST Global,General Catalyst,Index Ventures2SiPearl$37MSeries A2025-07-08N/A Cathay Venture,EIC Fund,France 20303Arago$26MSeed VC2025-07-07$60MEarlybird Venture,Protagonist,Visionaries Tomorrow,Acequia Capital,C4 Ventures4One Biosciences$17MSeries A2025-07-17N/A Blast,Redmile Group,Sofinnova Partners,Galion.exe,Invus Group5ArcaScience$7MSeries A2025-09-03N/AThe Moon Venture,Plug and Play Ventures,Akka Technologies,Bpifrance,Pleiade Venture 2025 CB Insights|141Israel142Quarterly funding&dealsState of AI|Geographic Trends|Israel$0.9B$0.5B$0.7B$1.2B$1.0B$0.9B$0.9B$0.5B$0.2B$0.3B$0.4B$0.2B$0.3B$0.5B$0.5B$0.3B$0.4B$1.0BFunding$0.6B432930506035354038292320172231232428Deals2801020304050607080$0.0B$0.2B$0.4B$0.6B$0.8B$1.0B$1.2BQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|143Annual percent of deals by deal stageState of AI|Geographic Trends|Israel62rvrrly-stage,65(%Mid-stage,21%9%9%7%8%Late-stage,9%Other,5 212022202320242025YTD 2025 CB Insights|144Israel:Top equity deals in Q325State of AI|Geographic Trends|IsraelCompanyRound AmountRoundDateRound Valuation Select Investors1Noma Security$100MSeries B2025-07-31N/A Evolution Equity Partners,Ballistic Ventures,Glilot Capital Partners2Exodigo$96MSeries B2025-07-16$700MGreenfield Partners,Zeev Ventures,10D Ventures,Jibe Ventures,Square Peg Capital3Irregular$80MSeries A2025-09-17$450MRedpoint Ventures,Sequoia Capital,Swish Ventures,Assaf Rappaport,Ofir Ehrlich4proteanTecs$51MSeries D2025-09-09N/AIAG Capital Partners,Addition,Intel Capital,Koch Disruptive Technologies,Porsche Automobil Holding5Blink$50MSeries B2025-07-28N/AO.G.Venture Partners,Lightspeed Venture Partners,Hetz Ventures,Vertex Growth 2025 CB Insights|145LatAm Trends146Quarterly funding&dealsState of AI|Geographic Trends|LatAm Trends$15M$272M$587M$207M$48M$199M$47M$147M$14M$14M$22M$18M$22M$121M$70M$99M$59M$53MFunding$126M191627232924192818191514161732192020Deals190510152025303540$0M$100M$200M$300M$400M$500M$600MQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|147Annual percent of deals by deal stageState of AI|Geographic Trends|LatAm Trends81rly-stage,88%9%7%Mid-stage,3%6%Late-stage,3%6%Other,5 212022202320242025YTD 2025 CB Insights|148LatAm:Top equity deals in Q325State of AI|Geographic Trends|LatAm TrendsCompanyRound AmountRoundDateRound Valuation Select InvestorsCountry1Enter$38MSeries A2025-09-24$375M Founders Fund,Sequoia Capital,Atlantico,ONEVCBrazil2Arvo$20MSeries A2025-09-17N/A Base10 Partners,Kaszek Ventures,K50 Ventures,CanaryBrazil3Tako$19MSeries A2025-07-30N/A Andreessen Horowitz,Ribbit Capital,ONEVCBrazil4Lastro$16MSeries A2025-09-30N/A Prosus Ventures,1Sharpe Ventures,Canary VC,Endeavor Scale-Up Brazil,QED InvestorsBrazil5BotCity$12MSeries A2025-09-25N/A Four Rivers Group,Astella Investimentos,Upload Ventures,Y Combinator,FirestreakBrazil6CAF$9MSeries A2025-09-24N/A B3,L4VBBrazil7Ursula$4MSeed VC2025-07-03N/A Prosus Ventures,Norte Ventures,Alex Chung,Betaworks,Fred SeibertBrazil8Pipeimob$3MSeed VC2025-08-12N/A HeadlineBrazil9Liquid AI$2MSeed VC2025-08-08N/A SaaSholic,Crivo Ventures,Flourish Ventures,Honey Island CapitalBrazil10ChambasAI$1MSeed VC2025-09-23N/A NAZCAMexico10Hunty$1MSeed VC2025-08-01N/A Cometa,Kalei Ventures,MatterScale Ventures,Newtopia VCColombia10Velum Labs$1MConvertible Note2025-09-13N/A Y CombinatorChile10WorkAI$1MSeed2025-09-23N/A M2Digital,Marcelo SmarritoBrazil 2025 CB Insights|149Quarterly exitsState of AI|Geographic Trends|LatAm TrendsIPO,02214M&A,2SPAC,0Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|150LatAm:Top investors by company count in Q325State of AI|Geographic Trends|LatAm TrendsInvestorCompany Count Investor GroupCountry1Bossa Invest5VCBrazil2GRIDS Capital2VCBrazil2ONEVC2VCBrazil2Scale-Up Ventures2VCBrazil 2025 CB Insights|151Brazil152Quarterly funding&dealsState of AI|Geographic Trends|Brazil$7M$241M$282M$93M$34M$183M$38M$18M$7M$10M$15M$11M$14M$110M$52M$85M$54M$34MFunding$124M61116162016141110978131125131113Deals1505101520253035$0M$50M$100M$150M$200M$250MQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|153Annual percent of deals by deal stageState of AI|Geographic Trends|Brazil82yrly-stage,85%8%Mid-stage,5%6%Late-stage,5%9%Other,5 212022202320242025YTD 2025 CB Insights|154Brazil:Top equity deals in Q325State of AI|Geographic Trends|BrazilCompanyRound AmountRoundDateRound Valuation Select Investors1Enter$38MSeries A2025-09-24$375M Founders Fund,Sequoia Capital,Atlantico,ONEVC2Arvo$20MSeries A2025-09-17N/A Base10 Partners,Kaszek Ventures,K50 Ventures,Canary3Tako$19MSeries A2025-07-30N/A Andreessen Horowitz,Ribbit Capital,ONEVC4Lastro$16MSeries A2025-09-30N/AProsus Ventures,1Sharpe Ventures,Canary VC,Endeavor Scale-Up Brazil,QED Investors5BotCity$12MSeries A2025-09-25N/AFour Rivers Group,Astella Investimentos,Upload Ventures,Y Combinator,Firestreak 2025 CB Insights|155Oceania Trends156Quarterly funding&dealsState of AI|Geographic Trends|Oceania Trends$75M$201M$244M$186M$85M$69M$94M$122M$42M$106M$103M$90M$21M$439M$136M$160M$175M$326MFunding$93M14191417221913121814131281819152314Deals17051015202530$0M$50M$100M$150M$200M$250M$300M$350M$400M$450M$500MQ1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|157Annual percent of deals by deal stageState of AI|Geographic Trends|Oceania Trends73wrly-stage,81%7%Mid-stage,9%9%5%8%Late-stage,2%6%5%Other,7 212022202320242025YTD 2025 CB Insights|158Oceania:Top equity deals in Q325State of AI|Geographic Trends|Oceania TrendsCompanyRound AmountRoundDateRound Valuation Select InvestorsCountry1Lorikeet$35MSeries A2025-08-06N/A QED Investors,Blackbird Ventures,Skip Capital,Square Peg Capital,AirTree VenturesAustralia2Andromeda$15MSeries A2025-09-09$66M Forerunner Ventures,Artesian VC,Main Sequence,Rethink Impact,Visible VenturesAustralia3Starboard Maritime Intelligence$14MSeries A2025-09-08N/AAltered Capital,King River Capital,OIF Ventures,Icehouse Ventures,Whakatupu Aotearoa FoundationNew Zealand4Zypher Network$7MSeries A2025-07-02N/A Signum Capital,UOB Venture Management,CatcherVC,Cogitent Ventures,DWF Ventures Australia5Brainfish$6MSeed VC2025-07-08N/A Prosus Ventures,Macdoch Ventures,SurgeAustralia6H3D$4MSeries A2025-09-16N/A Significant Ventures,Swinburne University of Technology,Co:ActAustralia7Alloy$3MPre-Seed2025-09-23N/A Blackbird Ventures,AirTree Ventures,Skip Capital,XtalAustralia7Marloo$3MPre-Seed2025-09-09N/A Blackbird Ventures,Brand Fund 1,Co Ventures,Matt Leibowitz,Philip FierlingerNew Zealand9Apate$2MSeed VC2025-08-25N/A OIF Ventures,InvestibleAustralia10Hall$1MPre-Seed2025-07-28N/A Blackbird VenturesAustralia10HeartLab$1MAngel2025-07-21N/A David OuyangNew Zealand10Mogoplus$1MVenture Capital2025-07-04N/A New Model Venture CapitalAustralia10Voqo AI$1MPre-Seed2025-07-14N/A Blackbird Ventures,Yellow Brick Road,Mark Bouris,Startmate,UNSW FoundersAustralia 2025 CB Insights|159Quarterly exitsState of AI|Geographic Trends|Oceania TrendsIPO,04334M&A,1Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q320212022202320242025 2025 CB Insights|160Oceania:Top investors by company count in Q325State of AI|Geographic Trends|Oceania TrendsInvestorCompany Count Investor GroupCountry1Blackbird Ventures5VCAustralia2AirTree Ventures4VCAustralia3BHP Ventures2CVCAustralia3Co Ventures2VCAustralia3Co:Act2VCAustralia3Icehouse Ventures2VCNew Zealand3OIF Ventures2VCAustralia3Skip Capital2Asset/investment managementAustralia3Square Peg Capital2VCAustralia3Square Peg Capital2VCAustralia3W232CVCAustralia 2025 CB Insights|161Report MethodologyYou can download the underlying data found in this report here:https:/ you have questions about the definitions or methodological principles used,or if you feel that your firm has been underrepresented,please reach out to .What is included:Equity financings into private companies only.Funding rounds raised by public companies of any kind on any exchange(including Pink Sheets)are excluded from our numbers,even if they received investment from a venture firm.Only includes the investment made in the quarter for tranchedinvestments.If a company does a second closing of its Series B round for$5M and previously had closed$2M in a prior quarter,only the$5M is reflected.Round numbers reflect what has closed,not what is intended.If a company indicates the closing of$5M out of a desired raise of$15M,our numbers reflect only the amount which has closed.Only verifiable fundings are included.Fundings are verified via(1)various federal and state regulatory filings;(2)direct confirmation with firm or investor;(3)press release;or(4)credible media sources.Equity fundings to joint ventures and spinoffs/spinouts are included.Unicorn data includes private companies valued at$1B or more in the private markets globally,per the same 4 sources listed above and relied on for funding events,which include valuations disclosed in credible media sources.The list is maintained publicly and updated in real time at https:/ notes:Israel funding figures are classified in Europe;funding to Oceania and Africa is included in global figures but not spotlighted in this report.Rounds to private companies that may be majority-or minority-owned subsidiaries of other private companies.Valuation data includes estimates to calibrate median and average valuations based on current and previous quarter disclosed valuations gathered from the aforementioned foursources.The estimating method will control for the over-sampling of large rounds that are reported quickly versus a comparative lag in valuations obtained from other sources.Valuation data reflects post-money valuations.Exits include IPOs,SPACs,publicly announced M&A deals,and other liquidity events;only first exits are counted.Headquarters are determined by publicly available sources including company-owned websites and profiles,legal filings,and press releases.All figures in the report are in USD.US financing trends follow the combined statistical area(CSA)methodology.Silicon Valley refers to the San Jose-San Francisco-Oakland CSA.What is excluded:No contingent funding.If a company receives a commitment for$20M subject to hitting certain milestones but first gets$8M,only the$8M is included in our data.No business development/R&D arrangements,whether transferable into equity now,later,or never.If a company signs a$300M R&D partnership with a larger corporation,this is not equity financing nor is it from venture capital firms.As a result,it is not included.No buyouts,consolidations,or recapitalizations.All three of these transaction types are commonly employed by private equity firms and are tracked by CB Insights.However,they are excluded for the purposes of this report.No private placements.These investments,also known as PIPEs(Private Investment in Public Equities),are not included even if made by a venture capital firm.No debt/loans of any kind(except convertible notes).Venture debt or any kind of debt/loan issued to emerging,startup companies,even if included as an additional part of an equity financing,is not included.If a company receives$3M with$2M from venture investors and$1M in debt,only the$2M is included in these statistics.No non-equity government funding.Grants or loans by the federal government,state agencies,or public-private partnerships to emerging,startup companies are not included.No fundings to subsidiaries of a larger parent corporation unless that subsidiary is a private entity and meets other criteria for inclusion.Accelerators,incubators,business-plan competitions,economic-development entities are excluded from rankings of most active investors,even if making equity financings.Rankings for top investors are calculated according to“company count,”or the number of unique companies an investor funds in a quarter,and so excludes follow-on deals.2025 CB Insights|162State of AI
2025-11-03
162页
5星级
Ahead ofthe game.Why its time for IP to think bigger with AI CONTENTS2AHEAD OF THE GAMEContents.010203Welcome to the future of IP.IP cant get no satisfaction.Its GenAI.What did you expect?040506Is there anything AI cant do?Its IP.But not as you know it.Say hello to AI you can rely on.Welcome to the future of IP.The great AI experiment is revolutionizing business whilst colliding with real-world economic pressures.But when it comes to the global intellectual property(IP)sector,wheres it going right?Wheres it going wrong?And wheres it going next?Our new research asked the questions.This report reveals the answers.And some just might surprise you.Todays hype focuses on the biggest foundational Large Language Models(LLMs)from the likes of OpenAI,Anthropic,Google and Meta.Our findings,instead,highlight an opportunity for IP professionals to move beyond these multi-purpose models.Those who do will find their need for trustworthy outputs met by smaller,bespoke models trained on quality industry data.Accuracy.Precision.Reliability.Security.Whatever tools emerge,these qualities are non-negotiables for practising IP law.Many externally characterize the IP sector as overly conservative and risk averse.Our research tells a different story.Showcasing an industry thats embracing innovation without compromising its integrity.We also challenge the profession to think bigger.Sure,GenAI solutions offer ways to work more efficiently and save costs.But the real rewards are found beyond the comfort of automation and assistance,toward new horizons of enhancement and transformation.But heres our take on it.GenAI is best explored in tandem with other technologies not in isolation.Why?Because GenAI brings particular risks.These risks are reduced by building solutions and transforming processes in an integrated way,drawing on decades of domain-focused AI development and expertise.And heres the tension.While GenAI is on the rise,many professionals told us that long-standing IP management systems(IPMS)just arent keeping pace.Respondents pointed to limited innovation,poor flexibility and frustrating integration with other business systems.And now?A growing number of dissatisfied users are hitting pause waiting to see what AI and machine learning can really do before they commit to changing platforms.Lets be honest,its still early days.But expectations are high,and the reality so far?Our research shows just how underwhelmed the IP sector has been with AI solutions.Far more users report low satisfaction than success.Its time to change that.That gap between promise and delivery isnt surprising.Over the past two years,weve seen a rush of over-hyped GenAI tools flooding the market-big claims,inconsistent results and more than a little AI-washing.But now,the dust is settling.The industry is regaining its balance1,moving from initial excitement into a more stable and strategic innovation cycle.WELCOME TO THE FUTURE OF IP3AHEAD OF THE GAME1-Post-GenAI hypeBut many AI use cases will always need input or oversight from human expertsIP professionals expect AI use cases to deliver value in the next two-to-five yearsThe sector will only adopt AI if it can trust GenAI outputs are accurate and reliableWhats next?WELCOME TO THE FUTURE OF IP4Our research highlights some clear takeaways:Most respondents expect AI to assist them with their current workloads.A few expect AI to go further and automate their workloads.Only a small minority are looking beyond assistance and automation to AIs transformative potential for creating new business models and revenue streams.We hope you find this report valuable.Got comments or questions?Wed love to hear them.Get in touch with us at 0201035AHEAD OF THE GAME WELCOME TO THE FUTURE OF IPIn with the old.In with the new.Before we dive in,lets distinguish two broad schools of AI.Traditional or classical AI This ones been evolving for decades.Quietly adopted across industries and use cases.2 Built on machine learning3 and predictive analytics4.Generative AI(GenAI)This one hit the big time in 2022 with the launch of ChatGPT.Focused on generating new data(text,code,images,etc.)based on probabilities.50102Today,which one of these schools is attracting the lions share of investment dollars?You guessed it:GenAI.Overwhelmingly.And it makes sense new technologies can only evolve at the speed of investment.2-Top 13 Machine Learning Techniques3-3 Types of Machine Learning4-10 Machine Learning Algorithms5-What Is Generative AI?5AHEAD OF THE GAME6AHEAD OF THE GAME WELCOME TO THE FUTURE OF IPMeet our research audience.Our research investigates IP technologies and AI.What are people in the industry thinking about them?What do they expect from them?And how are they using them?We ran an international survey across 33 markets.Half of respondents were from China,Japan and Korea the North East Asian powerhouse economies responsible for almost two-thirds of global patent applications in 20246.92%of respondents are(at least)intending to explore AI applications.71%are engaged with AI opportunities.55%have already trialed or implemented AI for at least one use case.Its fair to say were hearing from an informed audience who can speak credibly into the future of AI in the IP sector.The stats tell an interesting story about early movers.68%of European respondents have tested at least one GenAI use case.Theyre followed by 52%in North East Asia,44%in the Americas and 43%in Australasia.312Global IP professionals completed our online survey.Methodology196Respondents from corporate IP teams.4Survey translated into four local languages.Depth interviewsWith subject matter experts helped put the research results in context.Survey fieldwork March to April 2025.Special thanks to Gwilym Roberts of Kilburn&Strode.6-World Intellectual Property Indicators 20246AHEAD OF THE GAME116Respondents from IP law firms.Please note that not all chart totals will add up to exactly 100%due to data rounding.WELCOME TO THE FUTURE OF IP7AHEAD OF THE GAMEEvery number tells a storyTrends in IP practice.of IP law firms say licensing and monetization is definitely a growing part of their businessof corporates are monitoring for IP infringements,with 24tively converting to licenseesof corporate IP teams are prioritizing patent litigation over prosecutionof IP law firms are seeing more growth in patent litigation than patent prosecutionof corporate IP professionals are facing pressure to grow IP revenues,cut costs or both16R%7%Here are some of the stand-out statistics from our research.The headlines?Most IP professionals are feeling pressure to grow revenues but not doing everything in their power to make this happen.Many are taking small steps forward with AI but need to overcome significant barriers before leaping further.WELCOME TO THE FUTURE OF IP8AHEAD OF THE GAMEEvery number tells a storyThe AI journey.of IP professionals are attracted by the promise of AI to enhance efficiencies66%of IP professionals say accuracy and reliability is the#1 barrier to adopting GenAI 79%of IP professionals list security and data protection as the#2 barrier to adopting GenAI 62%of IP professionals have tested,trialed or implemented at least one GenAI solution55%of IP professionals expect AI to assist 50% of their current workload within five years 56%IP cant get no satisfaction.9AHEAD OF THE GAME02 10AHEAD OF THE GAME IP CANT GET NO SATISFACTIONMost of our corporate IP respondents work in small teams up to nine people.Half of these teams work as an independent business function.The other half sit within a larger corporate/legal or R&D/innovation function.All of them are busy.These,mostly small,IP teams manage portfolios of thousands,if not tens of thousands,of patents.They rely on an ecosystem of expert partners to get everything done both IP service providers and IP law firms.In fact,30%of corporate IP teams have completely outsourced their patent prosecution and litigation work.This figure jumps to 50%who are outsourcing at least three quarters of their patent activity.For IP teams to unlock the time savings promised by outsourcing,stakeholder communications need to be agile and secure.Yet only 40%of respondents are satisfied with how their comms are working.And only 6%are extremely satisfied.All bark.No bite?Its a tough economic climate out there.No wonder only 14%of corporate respondents say theyve escaped pressures to grow IP revenues or reduce portfolio management costs.Business priorities for corporate IP teams in 2025Base 196 corporate IP professionalsGrowing IP portfolio revenues26%Cutting IP management costs22%Both31%Neither14%Not sure6AHEAD OF THE GAMEOver a third of corporate IP respondents say theyve grown their headcount,service provider partnerships and law firm relationships over the last five years.By contrast,almost twice as many(62%)IP law firm respondents say theyve upped their staffing over the same period.With all this pressure on growth,its surprising that corporate IP teams arent tapping all available revenue streams.The majority say theyre actively monitoring for IP infringements(including a quarter who are very active).But far fewer are actively litigating or licensing.From an IP law firm perspective,only 16%say patent litigation is a bigger growth area than patent prosecution.Meanwhile,the same number definitely agree that IP licensing and monetization is a growing part of their business.Somethings amiss here.Otherwise,why the disconnect?Why invest scarce resources into monitoring for IP infringements only not to take action?Anthony Brennand,VP Product and Innovation,RWS IP Solutions,offers a clue:many existing patents arent robust enough to give businesses the litigation or licensing outcomes theyre looking for.This only heightens the pressure and scrutiny on IP teams from within their parent organization.IP CANT GET NO SATISFACTIONof Chinese corporate IP respondents are prioritizing growth in IP revenues(but theyre no more likely to be active in litigation or licensing than the global audience)42%Where are IP teams active?Base 196 corporate IP professionalsMonitoring for global IP infringementsLitigating infringementsConverting infringers into licensees2329241591225252513618223222Very activeSomewhat activeA littleNot at allUnsure12AHEAD OF THE GAMEOur findings also reveal a shift away from expensive,time-consuming IP protection work like patent prosecution.Over a quarter of law firm respondents say their trade secret work is growing,while only 17%are seeing growth in their copyright business.For Gwilym Roberts,Chair&Senior Partner at Kilburn&Strode,these questions of expense and time are central to the litigation question“People only really protect their IP when theres an infringement thats losing them sales.They then need a valid patent The cost and,vitally,disruption to management need to be proportionate.So,its a big deal there is only a handful of decided patent cases every year in the UK,for example”.Tech talk.Time for an upgrade?For decades,the IP Management System(IPMS)has been the go-to technology for IP professionals.But now they face the same existential crisis as Enterprise Resource Planning(ERP)and other established systems.Are they the best solution for todays fast-changing world?Our research suggests not.Take docketing,for example.Pivotal to managing large IP portfolios.But complex and time-consuming to complete.When 25%of IP professionals say one of the main attractions of AI is its ability to handle admin tasks,you can be sure most of them have docketing in mind.60%of our respondent organizations use an IPMS today.Yet only half are satisfied with its performance(falling to 12%who are extremely satisfied).Digging a little deeper,we see that even at these low levels,corporate IP professionals are 40%less likely to be extremely satisfied than their IP Law Firm peers.IP CANT GET NO SATISFACTIONOnly 44%of Chinese IP professionals are using an IPMS today44%PatentIP Law FirmsIP categories seeing growth areas in demandBase 116 IP law firm professionalsTrademarkCopyrightDesignTrade secrets661245151751127334GrowthclineAHEAD OF THE GAMEIP law firmSo,which aspects of the IPMS are still holding up?And which are crumbling down?Lets start with the positives.IPMSs are seen as robust.Reliable.Secure.Protective of data.Again,like on-premise ERPs.These are vital qualities of any system in the IP world.Within a broader operating environment where trust is low and risks are high,that gives IPMS vendors a real advantage.But where these systems are falling short matters just as much and in some cases,its actively blocking the integration of GenAI into day-to-day operations.Systems integration tops that list of downsides.Then user experience,flexibility and modularity,quality of reporting and data,and breadth of functions are the other attributes scoring poorly on satisfaction.Corporate IP respondents are less satisfied than their law firm peers with IPMS across all attributes except security and data protection.For example,only 5%are extremely satisfied with their IPMSs integration with other systems,with only 6%extremely satisfied with its innovation and development.In reality,enterprise plans to roll out GenAI at scale can quickly stall especially when faced with siloed IPMS platforms and the wider integration of IP data.Its no surprise then that the two words most commonly used by respondents to describe todays IP tech landscape were expensive(33%)and complex(32%).And heres the big one.A large minority of users(including over 40%of corporate IP respondents)think their IPMS will soon be obsolete.Not something youd generally hear said about an ERP.IP CANT GET NO SATISFACTION13Satisfaction with different aspects of current IPMS(%)Base 312 IP professionals(196 corporate|116 IP law firms)Innovation and developmentFlexibility and modularitySystems integrationReporting and data qualityRange of functionsReliability or robustnessSecurity and data protectionUX/CXExtremely SatisfiedExtremely SatisfiedSomewhat SatisfiedSomewhat SatisfiedCorporate IP6959111830231120132022281111AHEAD OF THE GAME14AHEAD OF THE GAMEOur research shows dissatisfied IPMS users are largely not migrating to new systems but are playing a waiting game.Why?Because theyre waiting for something better.They have a strong appetite to integrate IP data deeper into the enterprise,moving away from standalone systems.Even more corporate respondents are waiting to see how machine learning and AI plays out before making their next move.With so many IP use cases emerging,the roles of both IP paralegals and attorneys could look very different in the years ahead.Some tasks will be automated.Others will be assisted by AI.And new responsibilities will likely emerge as technologies become more deeply integrated.Thats the broader context in which we need to think about the future of the IPMS and of IP tech more generally.It also points to a growing need for new skills and expertise across the profession.When we zoom out to consider the ripple effects of AI adoption across wider corporate functions,its easy to see why many corporate respondents are holding off on IPMS decisions.The risk is that todays systems may not just be outdated,they may be out of place in the organizations of the future.IP CANT GET NO SATISFACTIONWhat comes after the current IPMS?Base:65 corporate IP professionals dissatisfied with their IPMSWe want to understand the impact of Machine Learning/AI on IP processes before considering IPMS optionsBase:35 IP law firm professionals dissatisfied with their IPMSWe want to move beyond standalone or siloed IP systems and to integrate IP data across our organizationStrongly agreeSomewhat agreeNeither/NorSomewhat disagreeStrongly disagreeStrongly agreeSomewhat agreeNeither/NorSomewhat disagreeStrongly disagreeCorporate IPIP law firmCorporate IPIP law firmIts GenAI.What did you expect?15AHEAD OF THE GAME03 16AHEAD OF THE GAME ITS GENAI.WHAT DID YOU EXPECT?If IP stakeholder communications and IMPS technologies are missing the mark,what comes next?All eyes are watching where GenAI goes from here.Ready or not,here AI comesOver 90%of respondents are at least aware(intending to explore potential applications)of AI.37%are at the discovery stage.34%have already moved into active,operational or systemic AI use.Only 8scribe themselves as inactive(not interested and no plans)with AI and LLMs.of Japanese corporate IP respondents describe themselves as inactive in relation to AI,LLMs and apps like Chat GPT compared to 8%of the global audience18%Engagement with AI,LLMs and apps like ChatGPTBase:312 IP professionals(196 corporate|116 IP law firms)InactiveAwareDiscoveringActiveOperationalSystemicCorporate IPCorporate IPCorporate IPCorporate IPCorporate IPCorporate IPIP law firmIP law firmIP law firmIP law firmIP law firmIP law firm10 B%8%1%5%1%1AHEAD OF THE GAMEIn a legal profession often,perhaps unfairly,characterized as conservative and risk averse,whats driving all this experimentation with GenAI?As it turns out,the answer has a lot to do with generic,short-term benefits.Efficiency,productivity,process automation and cost savings topped our list.This isnt surprising,given the commercial pressures corporate IP teams are under.But it also raises a bigger question:what does being conservative actually look like in a post-ChatGPT world?Where almost 70%of IP law firms have trialed unproven technology knowing that it will hallucinate(generate made-up or untrue responses).Where even the top-performing foundation models are wrong at least 15%of the time7.These same professionals are saying the very systems that are robust,reliable,secure and protective of data are no longer fit for the job.These voices dont sound conservative reflecting the impact of GenAI on yesterdays certainties.7-AI HallucinationBiggest attractions of AI solutions(%)Corporate IP base:171 corporate IP professionals aware of AIHow many have tested or trialed GenAI?Base:312 global IP professionalsCorporate IP52%IP law firm61%IP law firm base:102 IP law firm professionals aware of AIIncreased efficiencyBetter process automationIncreased productivityHandling admin tasksCost savingsAutonomous decision makingBreadth of capabilitiesContinuous improvementBetter personalizationOtherNot sure64685351425329202837231918121821131311711Corporate IPIP law firm ITS GENAI.WHAT DID YOU EXPECT?18AHEAD OF THE GAMEFor Gwilym Roberts at Kilburn&Strode,the conservatives are ahead of the technology at the moment“we completely accept that AI is going to change things,and we really would love to embrace it,but it doesnt seem ready yet”.Not everyones being won over so quickly,though.Theres still a subset of legal professionals that remains fiercely opposed to GenAIs advances.Back in 2023,research found a quarter of legal respondents saying GenAI shouldnt be used for legal work(an ethical judgement).Even our own broadly supportive respondents had some negative things to say.This visceral reaction is partly driven by fear.Of change.Of disruption.Of loss to status,income or employment.Loss aversion is a powerful opponent to change,especially where people dont see any positive trade-offs.AI wont replace IP professionals any time soon.But like in other industries it will limit future opportunities for those without AI experience and skills.So,how can businesses and law firms address such fears around AI?Open conversation goes a long way.Talk about whats changing,why and how.How might peoples work change?What education and training will they get?Whats the longer-term vision?Change is on the horizonTalk about AI in organizations often revolves around the day-to-day.Theres not so much airtime for considering the technologys transformative potential.This is just like the adage goes:people tend to overestimate what technology can do in the short term,but underestimate what it can do in the long term.Our own research echoes this.When asked about different horizons for AI,only a minority of respondents say their organizations are thinking about enhancement.By contrast,60%are thinking about AI for automation.And almost 70%for assistance.Of course,automation and assistance are more immediate,tangible and relatable for people.Future thinking is hard.AutomationNo AI aspirationsNot sureWhats your current ambition for AI?Base:312 IP Professionals(196 Corporate|116 IP Law Firms)5%Business IPIP law firmAssistanceEnhancement59apeEH%6%7%ITS GENAI.WHAT DID YOU EXPECT?19AHEAD OF THE GAMEWhat about the question of enhancement?By this we mean reimagining business operations.Enhancing processes,products,services and revenues.Or even creating new ones altogether.As we settle into the AI age,well see lots more of this.Just as we hear about AI birthing new types of jobs,this decade of disruption will also bring fresh business models and organizational structures to life.But first,lets address the elephant in the room:how can the IP industry trust AI with business-critical work?The latest research shows even the best GenAI model has a 15%hallucination rate.As James Lacey,Senior Vice President,RWS IP Solutions,makes clear,“thats not good enough for the IP market.A 15%error rate would be disastrous for patent applications where even the slightest of errors can alter the entire scope of patent protection.”Its hard to see how GenAI alone can be trusted to work autonomously.If enhancements the end goal,we first need to fix the reliability problem.To build or buy?That is the question.We asked IP professionals how they expect to implement GenAI.Buying it.Or building it.Most believe theyll do both.This suggests a healthy awareness of the risks of GenAI,with potential for a spectrum approach.Lower-value applications can be bought off the shelf,running autonomously in the public cloud.Higher-value applications can be developed,trained,fine-tuned and maintained with more investment to reduce the risks of hallucination.Ultimately,as Anthony Brennand says,“Once were through the experimentation phase,many organisations will rely on AI powered applications,where the vendor is responsible for managing security risks,keeping up to date with the latest models and the expense of maintaining an AI solution.”Are you interested in building or buying GenAI solutions?Base:286 IP professionals(182 corporate|104 IP law firms)BuildingBuying18%NeitherNot sure9%9%Corporate IPIP law firmBoth29)%ITS GENAI.WHAT DID YOU EXPECT?Is there anything AI cant do?20AHEAD OF THE GAME04 21AHEAD OF THE GAME IS THERE ANYTHING AI CANT DO?We know that 55%of organizations we surveyed have tested,trialed or implemented at least one GenAI use case already.Its actually 3.5 use cases,on average.But which are the most common IP use cases?And how satisfied are IP professionals with the results?In an AI market buzzing with outlandish claims of cost savings and efficiencies,these are the questions that need answering.Gwilym Roberts at Kilburn&Strode,for one,is seeing“a lot of solutions in search of a problem in the sense of people saying,weve got AI that does this we have no idea if its relevant to you,but do you want to buy it?.Nobody is really asking people what they want from an AI offering.”Topping the use case charts is patent translation.Then,rounding out the top five,we have patentability searches,patent drafting,report drafting and dedicated IP advice chatbots.GenAI use cases trialed or implemented(%)Base:103 corporate IP professionals who have trialed at least one GenAI solutionCorporate IP professionalsIP law firm professionalsImage search(trademarks/design)Patent database optimizationPatent draftingReport draftingPatent translationPatentability,FTO,invalidity,SOA searchAdvice chatbots for IPPatent/TM prosecution and PTO interactionsImage search(patent)Identifying licensing opportunitiesBase:65 IP law firm professionals who have trialed at least one GenAI solutionTotalTrialed221626213936443866515543342820123326161437282522624855427150644747355035393015822AHEAD OF THE GAMEPatent translation gets the highest satisfaction scores among IP professionals.It also gets the lowest.Reflecting the fact that its the use case most organizations have experience with.Only a minority are highly satisfied with AI for patent translation 27%among corporates,38%among law firms.Most respondents are ambivalent.Thats more than we can say AI patent drafting.This scored rock bottom for satisfaction,especially among corporate IP teams.Subtlety.Nuance.Precision.Patent drafting needs all three.It seems todays AI tools just arent up to the task.But that can change.80%of our respondents still expect AI patent drafting to deliver on the promise within five years.Only 9%think it never will.Is that too optimistic?Maybe not,given the pace of progress with AI.Even a few short years ago,not many people could have imagined what GenAI applications can now do across code,prose,images,statistics,video and voice.Thats the power of machine learning.IS THERE ANYTHING AI CANT DO?of IP professionals in China think AI patent translation will realize its value within two years,compared to 71%in Japan85%Satisfaction with GenAI use cases trialed or implemented(%)Base:312 IP professionals(196 corporate|116 IP law firms)Corporate IP professionalsIP law firm professionalsImage search(trademarks/design)Patent database optimizationPatent draftingReport draftingPatent translationPatentability,FTO,invalidity,SOA searchAdvice chatbots for IPPatent/TM prosecution and PTO interactionsImage search(patent)Identifying licensing opportunities29142895293117132731152920172242830691716536141118113824101919319274820LowHighBase:103 corporate IP professionals whove trialed at least one GenAI solutionBase:65 IP law firm professionals whove trialed at least one GenAI solution23AHEAD OF THE GAMEEarlier we saw that the industry isnt licensing and monetizing IP as much as it could.Our research shows some respondents have trialed or implemented AI to identify IP licensing opportunities.But the results have been underwhelming.Again,its the hurdles of subtlety,nuance and precision that AI just cant seem to clear.This is also apparent in image search use cases for patents,trademarks and designs.Typically,our respondents have looked at 3.5 AI use cases but implemented just one.AI vendors would hope for more.But its a fair reflection on a new technology still in its experimental phase.For IP teams,a tipping point in AI adoption seems just around the corner.But which use cases should come first?And how dependable are they?IS THERE ANYTHING AI CANT DO?23AHEAD OF THE GAMEHow many GenAI use cases trialed and implemented(on average)?Base:103 corporate IP professionals and 65 IP law firm professionals who have trialled at least one GenAI solutionCorporate IP3.5Tests/trials0.9ImplementationsIP law firm3.3Tests/trials1.2Implementationsthe volume of GenAI solution trials than implementaions3x24AHEAD OF THE GAMEYou couldnt make this up(but AI could)We know around 90%of IP professionals expect AI to deliver value across use cases.And we know theres mounting pressure on corporate IP teams to do more with less.Chances are,autonomous AI is on the way.That would align with the current talk right now about Agentic AI8.The next big thing.The idea is AI agents will complete complex tasks independently on your behalf.Interacting with multiple service providers to get the best results for your personal needs.In practice,most of our respondents are haunted by AI hallucinations and the harm they might cause.Across 11 AI use cases,45%of IP professionals(on average)believe therell always be a need for expert human oversight.For Gwilym Roberts at Kilburn&Strode“we will certainly not be in a position,if only for insurance reasons,to simply sit back and hand it over to AI for some time to come”.Expert human-in-the-loop9 models are a game changer for user confidence.And,so,for results with AI solutions.In 2023,driverless car operator,Cruise,was caught exaggerating its AI capabilities.In reality,humans were still behind the wheel of critical tasks10.Great for road safety.For AI hype and share prices,not so much.But do human-in-the-loop models really count as wins for AI?It depends who you ask.For those obsessed with academic AI or pursuing Artificial General Intelligence(AGI),probably not.But for business users who care only about trustworthy outputs to make sound decisions affecting thousands of people and the bottom line its a resounding yes.The better question then is not whether they are real AI,but real solutions.Because in a field as technical,precise and detailed as IP,its real solutions we need.Patent translation stands out as the most widely used and most satisfying AI use case in IP today.Even so,45%of respondents believe expert human oversight will always be essential.Thats no surprise.The human-in-the-loop approach is already a core part of OECD recommendations on AI11.The need for human oversight,or more accurately,the importance we attach to quality,will temper the ability to drive efficiencies with AI.In IP,precision and accuracy arent just preferences.Theyre critical to getting things right.As we move beyond the initial GenAI shockwave,these fundamentals will be essential to building smarter,more reliable solutions and models.IS THERE ANYTHING AI CANT DO?9-What is human-in-the-loop?10-Cruise confirms robotaxis rely on human assistance11-Respect for the rule of lawHow many IP professionals think these AI use cases will always need human oversight?(%)Base:182 corporate IP professionals and 104 IP law firm professionalsPatent/TM prosecution(PTO interactions)PatenttranslationPatentdraftingPatentdatabase optimizationPatentability searches(FTO SOA etc.)Dedicated IP advice chatbotsIdentifying licensing opportunitiesImage search(patent)Image search(trademarks and design)Patent/TM/Design renewals and recordsReport drafting557035456786Corporate IPIP law firm256260694233376159243120201331458-What Is Agentic AI?The adventures of Copilot and Claude25Lets zoom into a couple of leading AI tools gaining traction in the workplace-Microsoft Copilot12 and Claude13 from Anthropic-with insights from two recent studies.Software development and technical writing are the two biggest use cases for Claude.People tend to use it to augment their work(57%)rather than automate it.Anthropic found that just 4%of workers were using Claude to support 75%or more of their work tasks compared to 36%using it for 25%or less.The people using Claude the most tend to be in mid-salary roles(computer programmers,copywriters,etc.).Highest and lowest earners arent jumping in yet.Microsoft Copilot is a different kind of model.Bundled together with Microsoft 365,its already within clicking distance of a vast universe of users.The research shows 40%of workers with access to Copilot are using it regularly.The main takeaway?Copilot is saving people small but significant amounts of time when using Microsoft Teams,Outlook and Office.In other words,less time reading emails.Less time creating documents.Less time in meetings.More time for everything else.These are positive results.In fact,80%of workers wanted to keep using Copilot after the six-month study finished14.IS THERE ANYTHING AI CANT DO?12-Early Impacts of M365 Copilot13-The Anthropic Economic Index14-What Can Copilots Earliest Users Teach Us About Generative AI at Work?AHEAD OF THE GAMEIts IP.But notas you know it.26AHEAD OF THE GAME05 27AHEAD OF THE GAME ITS IP.BUT NOT AS YOU KNOW IT.As AI capabilities accelerate,every industry is facing big questions.How will we choose to deploy these tools?Will businesses lean hard into automation,speed and efficiency?Or will they lead with their human edge prioritizing precision,quality and trust?Its no surprise that interaction of rapid innovation is leading to real-world financial pressure.That combination naturally pushes businesses to focus on short-term efficiencies and cost savings.The real winners will be the IP teams that think bigger.Exploring how AI can enhance and transform business models.Solve new problems.Generate new revenues.Create new roles.Were talking about one of the most powerful and game-changing technologies humanity has ever created.Surely theres more to gain than just saving a few minutes or dollars while business ticks over as usual?AI-driven efficiencies are already changing the shape of the workforce and that shift is only set to grow.The real question is:what new opportunities will emerge as we reshape operations and reimagine entire industries to make space for this new technology?27AHEAD OF THE GAME28AHEAD OF THE GAMEThe future of workWhat happens to the shape of IP work in the next five years?We asked respondents how they see their workload evolving specifically,how much they expect to be automated,and how much they think will be assisted by AI.ITS IP.BUT NOT AS YOU KNOW IT.How much of your IP work will AI assist or automate?Base:312 IP professionals|196 corporate and 116 IP law firmsCorporate IP professionals100pP0 %0%Not sure0 5 10 15 20 25 30IP law firm professionals0 5 10 15 20 25 30AssisstedAutomated%workload%respondent29AHEAD OF THE GAMEof IP work will be automated by AI20-30%Most IP professionals expect AI to automate 1030%of their current work.More broadly,almost one-in-five predict that AI will assist with three quarters or more of their role.What could this mean for employment in the IP sector?Based on current workloads and survey responses,we can start to sketch a possible future:of IP work will be assisted by AI40-60%of IP work will be done by humans 20-30%ITS IP.BUT NOT AS YOU KNOW IT.29AHEAD OF THE GAME30AHEAD OF THE GAMEThe implications for jobs are clear.How big will these changes feel?How fast will they happen?This will vary by organization and role.But one things certain:IP professionals need to focus not just on growing the value of their uniquely human skills,but also on learning how to get the most out of AI tools.The future of the business modelAs AI solutions boost efficiency,most IP professionals expect that more IP work will be brought(back)in-house.Its a view shared by corporate and law firm respondents alike.This confidence is brittle,though.Only 14%see in-housing as extremely likely.While the profession believes the promise of things to come,its experiences with use cases and solutions so far show this is more aspiration than reality.But heres the worry.IP is a complex field for AI to handle.IP teams are already under-resourced and over-stretched.Forcing them to adopt immature tools that lack the quality needed could actually harm efficiency not help it.Without expert human-in-the-loop solutions,sophisticated use cases will take a huge manual effort from teams to make sure the outputs are accurate,precise and consistent.15Employee morale and performance quality also plummet when organizations rush to implement AI without a clear purpose,strategy and co-ordinated approach.16Is this where IP service providers are stepping up?You might think so.But our research suggests otherwise.Unfortunately,only 20%of respondents are satisfied with the AI expertise and guidance from their IP service provider.This needs to change.Fast.Our respondents expect most AI use cases to be delivering real value within two years.When this happens,the in-housing aspiration will turn to reality.IP service providers without AI expertise will get left behind.What does this AI expertise need to look like?Well,that depends.Each provider should find their own sweet spot in the value chain.Some will be best suited to simple automation use cases.At RWS,we bring a broader,deeper understanding built on years of IP experience,AI innovation,process transformation,translation and localization expertise.We obsess over solving the most technical and demanding challenges across the IP lifecycle.Aspiring to deliver the best solutions not always the first ones so that you can get fit for the future and innovate with confidence.ITS IP.BUT NOT AS YOU KNOW IT.of IP professionals in the Americas say its likely AI efficiencies will move more IP work in-house,compared to 72%in China32%Will AI efficiency mean more work done in-house?Base:312 IP professionals|196 corporate and 116 IP law firms14%Extremely likelyCorporate IPIP law firm14C%Somewhat likely42)%Neither/nor35%9%Somewhat unlikely7%4%Extremely unlikely2-From Burnout to Balance16-AI Implementation Can Hurt Employee Engagement31AHEAD OF THE GAME06 Say hello to AI you can rely on.32AHEAD OF THE GAMEWhere does all this leave us?Lets recap the headlines so far.The IP sector is making significant steps forward with AI.But theres still a way to go before it realizes the benefits across the board.This isnt surprising.GenAI is still a new and experimental technology.As weve seen,even the top-performing model hallucinates 15%of the time.While the chances of other leading models making things up are no better than a coin toss17.No wonder so many IP professionals say accuracy and reliability are the biggest barriers to getting on board with AI.And 45%of our respondents say expert human oversight will always be needed across 11 IP use cases.Bottom line:the IP profession doesnt trust AI yet.This distrust is only fuelled by non-technical issues like concerns of copyright in training data collection.Legal professionals want to know theyre working with technology partners that take ethics and responsibility seriously.The sooner theres legal clarity here,the better for everyone.For IP teams,differentiating between vendors and propositions is easier said than done.Just look at the low satisfaction scores with AI use cases from our respondents.Efficiency.Automation.Savings.Productivity.These are grand promises.But too often the results dont follow.Making teams even more hesitant to commit to big investments.Are vendors to blame?Not entirely.After all,theres no plug-and-play formula for AI success and financial returns.But the statistics show a collaborative partnership between humans and machines stacks the odds well in your favor.When BCG published their analysis18 they highlighted the symbiotic relationship between AI and human intelligence and emphasised the importance of process transformation and organisational learning.Its worth revisiting now as we move toward a more thoughtful,deliberate phase of AI adoption.It also echoes RWSs own belief in Genuine Intelligence.What stands out in our research is the level of confidence IP professionals already place in AI.For all the talk of the sector being conservative and risk-averse,theres clear appetite to engage and strong belief in what the technology could become.That optimism may feel at odds with todays lived experience.But with AI systems continuously improving and learning,its not unrealistic to say:the best is still to come.SAY HELLO TO AI YOU CAN RELY ONThe IP sector is making significant steps forward with AI.But theres still a way to go before it realizes the benefits across the board.17-AI Hallucination18-Are You Making the Most of Your Relationship with AI?32AHEAD OF THE GAME33AHEAD OF THE GAMEAll ChangeAI is here to stay.The way we work will change.How much?How quickly?Honestly,its tough to tell.But there are some things we can predict with a little more certainty.First.Therell be a boom in high-quality small language models(SLMs)custom built for individual organizations,industries and even countries.Theyll hallucinate less.Run faster and cheaper.Use less energy.Keep data more secure.And protect reputations.Second.The industry will find smarter,cheaper and more efficient ways to deliver GenAI.Just as DeepSeek allegedly did the impossible by building a credible rival to ChatGPT at a fraction of the cost,so well continue to see innovation pick up pace.Third.Many of the best solutions will integrate GenAI with other technologies and expertise to improve performance.Fourth.While the pioneering EU AI Act is already having an impact,the global GenAI market is unlikely to see cohesive regulation any time soon.This will put a premium on organizations committed to strong governance and ethical AI thats fair,accountable,explainable and transparent.Wed love to see a race to the top with high-quality,responsible AI.Ready.Set.Go.What does this all mean for the IP sector?Good things.This second wave of more reliable GenAI solutions will be game changing for the legal profession.Organizations should get ready to capitalize by educating and training their people in AI skills.Service providers should adapt as more IP prosecution,litigation and licensing work moves in-house.Looking to the horizon of enhancement.Reimagining their businesses for the AI age,reflecting their expertise across the IP value chain.The IP profession will need to overcome legal and cultural barriers if its going to get comfortable with GenAI.Bespoke small language models are already gaining traction.If copyright is enforced,expect demand for these to shoot through the roof.How can the industry make sure disappointments today dont dampen innovation tomorrow?IP service providers and law firms have an opportunity here.Sharing lessons learned from AI experiments to help businesses avoid getting burned by immature or overblown solutions.This will pave the way for progress.As IP professionals raise their heads above immediate pressures.Embracing the many opportunities AI brings to grow the value of their work.As Gwilym Roberts of Kilburn&Strode puts it“there will come a point where it is negligent not to use AI,when it becomes better than people.and we need to be ready for that.”Its an exciting road ahead.If youre looking for a trustworthy AI partner to light the way,wed love to talk.SAY HELLO TO AI YOU CAN RELY ON33AHEAD OF THE GAME34AHEAD OF THE GAMESigns of Genuine Intelligence.34At RWS,we dont just have a 60-year heritage in IP language services.We also have more than two decades of experience at the cutting edge of AI(with 47 AI technology patents to show for it).As a result,our translation and communication solutions are uniquely built for the exacting demands of IP law.Since GenAI hit the mainstream,weve been advocates for Genuine IntelligenceTM.Its an approach that combines the best of human expertise with advanced IP-trained AI technology.Two intellects working as one each complementing the others strengths to give you more than the sum of their parts.Chasing shiny toys is one thing.Purposeful innovation?Thats quite another.And its what we stand for.Recent waves of new AI models,tests and products have had a dizzying effect on the market.Organizations have felt compelled to get their own AI story straight.Executives have felt the pressure to deliver on AI use cases(or at least be seen to).All the while,talk of existential risks,superintelligence and AGI have clouded legitimate concerns about AI performance.Our research backs this up.55%of IP professionals say their organizations are actively using GenAI.Whoever says the IP sector is afraid to roll with change is clearly living in the dark.Is the technology sector going to leave the IP profession burned by bad experiences with unfit and immature AI solutions?Not on our watch.SAY HELLO TO AI YOU CAN RELY ONAHEAD OF THE GAME35AHEAD OF THE GAMELets talk about you.Got questions about what youve read in this report?Thinking about your next steps with AI or IP services more broadly?Were here to help.SAY HELLO TO AI YOU CAN RELY ONEmail us:Our contact form: us online: OF THE GAME RWSAbout usRWS is a content solutions company,powered by technology and human expertise.We grow the value of ideas,data and content by making sure organizations are understood.Everywhere.Our proprietary technology,45 AI patents and human experts help organizations bring ideas to market faster,build deeper relationships across borders and cultures,and enter new markets with confidence growing their business and connecting them to a world of opportunities.Its why over 80 of the worlds top 100 brands trust RWS to drive innovation,inform decisions and shape brand experiences.With 60 global locations,across five continents,our teams work with businesses across almost all industries.Innovating since 1958,RWS is headquartered in the UK and publicly listed on AIM,the London Stock Exchange regulated market(RWS.L).More information:Copyright 2025 RWS Holdings Plc.All rights reserved.
2025-11-03
36页
5星级
Sovereign AI:Own your AI future From managing risk to accelerating growthCopyright 2025 Accenture.Al.
2025-11-03
24页
5星级
SURFTECH TRENDS2026The digital transformation is fundamentally affecting the way that participants in our society from citizens to companies and public organisations interact,collaborate,learn,and perform.Digital developments are accelerating,resulting in for example exceptionally steep adoption rates of artificial intelligence(AI)in recent years.This is also seen in AI integration in end-user applications(e.g.Copilot,WhatsApp),but also in other layers of the so-called digital technology stack,such as computing power,extended reality(XR),data networks,and data management.This evolving digital transformation is of paramount importance for the cooperative SURF as a collaboration of national research organisations and education institutes.For most members of SURF,it is a challenge to anticipate the potential added value,risks,and impact of emerging digital technologies and to manage topics like digital trust,digital wellbeing,and digital autonomy,in a responsible manner.Looking to the future horizons of digital developments is essential for SURF cooperative.It is important for the sector to harness emerging technologies for research and education,with public values as a foundation.Futuring,scenario planning,and technology exploration capabilities are needed.This to search these horizons,to identify whats likely to emerge and to identify the potential value and impact of digital technologies on research and education.SURFs futuring activities are supporting the cooperative to collectively learn and enhance our capabilities regarding our future-readiness.SURFs continuous technology exploration and knowledge development supports a deep understanding of challenges,risks and impact of emerging technologies for our sector.We are proud that this SURF Tech Trends Report is one of the activities and resources to collaborate and serve the members,and to give the various stakeholder groups in our cooperative insights to act.Hans Louwhoff and Ron Augustus (Board of Directors SURF)PrefaceSURF TECH TRENDS 2026Navigating through the stackContentsIntroductionApproachReading guideValue compassDriversSURF TECH TRENDS 2026Immersive TechnologiesDigital TrustConnectivityData ManagementCybersecurityQuantum TechnologiesWildcardsArtificial IntelligenceCloud ComputingComputingIn a fast-changing,dynamic environment,keeping up with the development can be a challenge.With the previous Tech Trends report(2023),we have made a start in supporting and facilitating the SURF cooperative by identifying and curating relevant trends for education and research.Compared to 2023,this edition is bigger,bolder and goes beyond.BiggerIn the production of this Tech Trends report,about thirty professionals from over twenty member institutions participated in the working groups together with SURF colleagues.Together,we have covered ten technology-oriented chapters that we present in this report.Besides the working groups we engaged with other stakeholders with knowledge and experience to contribute to the tech trends.BolderFor each trend,we have looked at its potential impact on education,research and operations.Operations is a new impact area that has been added as we see IT becoming a more integral part of an institution or campus.Along with the nine technology-driven chapters that are selected,we have added a chapter with wild cards that can either be black swans or grey rhinos due to their non-obvious and less predictable character.BeyondAs we have our biases as people working in education and research,we have consulted“outsiders”:industry leaders and experts.The perspectives we have included represent not only big industry but also European and more global view points.Throughout this report,you might come across the insights they shared with us.Some highlights to keep in mindOverarching in this report,a few observations have stood out that demand attention moving forwards:There is a significant impact of AI on the trends.Therefore we can conclude AI as a system technology There is a strong aim for efficiency(doing more with less)in technologies More attention needs to be paid to digital skills to navigate the digital transformation,especially in the spirit of cybersecurity There is a strong tendency to look only towards the USA,and we shouldnt overlook other countries like Singapore and China As a lot is happening,we see strategic options emerging to strengthen Netherlands position globally.Gl Akcaova(lead Futurist SURF)IntroductionSURF TECH TRENDS 2026Our project principles In collaboration with SURF members as it is primarily for the members IT-oriented view Outside-in perspective Trend curation rather than trend watching Readable(introductions)for board members and C-suite(actionable)insights for CIOs,information managers,policy advisors,etc.Impact assessments(opportunities and threats)for Education,Research&Operations,rather than defined or prescribed recommendations by SURF Conversation starter rather than a“marketing”report or an“how-to”Selection of the chaptersOver 23 technology reports are identified,scanned and clustered.Eventually,we have looked and identified the common denominators that led to the nine chapters.To ensure novelty and inspiration,an extra chapter(wildcards)is created.Selection of the trendsThe working groups were formed based on a call for expertise/contributions,which was widely spread.Based on expressed interest,we aimed for well-balanced groups per chapter.We had a kick-off on Oct 10,2024,to set the foundation to scan(literature review),validate(interviews and consultation),and analyse(group discussions)the trends as groups.On May 7th,a wide expert consultation was organised where a mixed group from the cooperative was invited to assess the recognisability and relevance of the selected trends by the working groups(see acknowledgement).Over the summer,the editors reviewed the content,assessed the quality and consulted reviewers to further improve and ensure accessibility of the content of this report.ApproachAI disclosureAI(LLMs)was used by various working groups to analyse documents,assess trends and improve the writing.AI was also used by the graphic designer for the chapters covers.v.1-2024-mayPROJECT INFOGRAPHICJANJAN20242025SEPTPlanningProject briefing&planJanuary-May 2024SelectingFirst selection of tech chaptersMarch-May 2024ScanningContextual factors of Education&ResearchMay September 2024FormingFormation of core teams October-December 2024PerformingTrend analysisJanuary-May 2025FinalisingFinal edits and lay-outMay-Augustus 2025PublishingDelivery of Tech Trends reportSURF Summit2025Road to Tech Trends ReportAfter a successful Tech Trends Report in 2023,SURF is about to start creating the trend report for 2025-2026!The SURF Tech Trends Report is a bi-annual report for the members within our education and research cooperative.It presents an overview of the most relevant technologies and helps our collaborative organisation with navigating the complexity of technological advancements.A lot is happening in a rapidly changing environment,and it is becoming a challenge to keep up.There are many reports out there,but none of them are tailored to education and research.Therefore,SURF curates and analyses what is shaping-or has the potential of shaping-the future of education and research.Some of our project principles For the members with the members Outside-in perspective Trend curation rather than trend watchingSURF TECH TRENDS REPORT 2025-2026Tech trends report 2023 more infoSURF TECH TRENDS 2026Reading guideIncreased application of FAIR principles to enable digital ecosystems in EuropeTREND#1The FAIR principles are guidelines on datasets being findable,accessible,interoperable,and reusable.Increased application of FAIR principles may lead to digital ecosystems,a web of FAIR data and services,where digital resources are not only shared but also semantically linked,automatically interpreted,and reliably reused across domains.Beyond data,a key evolution is the application of FAIR principles to all research outputs,including supporting frameworks such as software,computational workflows,and scientific models.This convergence is facilitated by FAIR Digital Objects(FDOs),which include persistent identifiers and rich metadata,alongside Knowledge Graphs that structure the semantic relationships between these FDOs.The shift to applying FAIR beyond just data only,to include digital research objects,enables automation,reproducibility,and information discovery,while also fostering cross-domain innovation on a larger scale.MaturityDrivers WATCH PLAN ACTPublic ValuesAutonomyJusticeIntegrity I Transparency|SustainabilityHumanityCommunity dynamics&social cohesion;Globalisation;Digital transformationSURF TECH TRENDS 2026Data ManagementInitiatives&organizationsGO FAIR Initiative(go-fair.org)#EOSC(European Open Science Cloud)(eosc.eu)#FDO Forum(fairdo.org)#OpenAIRE(explore.openaire.eu)#ImplementationsAridhia FAIR Data Services-provides researchers with tools for dataset discovery,classification,and metadata browsing()#Fairdata.fi-provides data storage and discovery services to support FAIR principles and ensure long-term preservation(fairdata.fi)#FAIRsharing.org-provides for a curated,informative resource on data and metadata standards,inter-related to databases and data policies(fairsharing.org)#Open Research Knowledge Graph(ORKG)-scholarly communication exploiting the possibilities of digitization (orkg.org)#Zenodo(zenodo.org)#FAIRimpact project(fair-impact.eu)#Supporting literatureToward the Open Science model:publish your raw diffraction data(pubs.aip.org)#Analysis on open data as a foundation for data-driven research()#Leiden Declaration on FAIR Digital Objects(fdo2022.org)#SIGNALS“The web of FAIR data and services is a keystone vision for the future of digital ecosystems.”-Paolo Manghi,OpenAIRE,CTOSURF TECH TRENDS 2026Data ManagementIMPACTEducation Knowledge Graphs(KGs)are being integrated into Learning Management Systems(LMSs)such as Brightspace and Canvas to model relationships between learning objectives,content,and competencies.Fostering links between learner profiles,learning progress,and course requirements,KGs enable personalised learning through adaptive learning paths.Research The web of FAIR data and services significantly impacts research by accelerating scientific discovery and promoting reusability.FAIR Digital Objects(FDOs),enriched with fine-grained metadata and versioning,enhance the reproducibility,traceability,and reuse of research data,software,and workflows.Operations Knowledge Graphs(KGs)can enhance operational systems by linking data from Learning Management Systems(LMSs)and Student Information Systems(SIS).The integration of KGs enables institutions to map the student journey from pre-admission to graduation and subsequent alumni engagement.SURF TECH TRENDS 2026Data ManagementChapterTrendReadiness levelDriversIntroduction of the trendSignals for this trendRead more about the signal onlineImpact of the trend on education,research and operationsNavigate through the pagesGo to the table of contentsTrend numberPublic valuesSURF TECH TRENDS 2026Value compassSURF and Kennisnet,the public IT organisations for education in the Netherlands,have developed the Value Compass to provide a common language to stimulate the dialogue about digital transformation in education and the importance of educational values.The Value Compass#provides a frame of reference for structuring digital transformation based on values.For each trend we identified a relationship with public values.The value compass helps us to identify a value or multiple values driving the trend towards a certain direction.foot-ball clubhumanity autonomyjusticeFreedom of choiceIndependencePrivacyPluralismAccountabilityIntegrityInclusionTransparencyEquitySustainabilitySocial cohesionRespect SafetyWell-beingMeaningful contactSelf-developmentSURF TECH TRENDS 2026DriversWe are living in an increasingly complex and interconnected world where powerful forces(drivers)are influencing and reshaping the foundations of society and,therefore,education and research.These drivers span social expectations,cutting-edge technologies,economic shifts,environmental urgency,and evolving policy and regulatory landscape(s).These drivers do not act in isolation;their convergence creates both disruption and opportunity,challenging traditional models of knowledge and institutional purpose,as the environment becomes uncertain and/or unpredictable.Understanding these drivers is essential not only to remain relevant but to make better decisions and prepare for a future that is already here.Economical drivers Global trade&tariffs Globalisation Concentration of wealth&economic inequality Service-oriented&value-based economies Digital transformationTechnological drivers Automation&AI Engineering advances&computation Biotechnology Connectivity&interaction Cybersecurity&trustEcological drivers Climate change&global warming Energy supply&demand Raw material scarcity Clean water demand Biodiversity Political and regulatory drivers Geopolitics&(digital)sovereignty Compliance®ulation Critical infrastructure Weaponisation of knowledge Ideologic polarisationSocietal and demographic drivers Individualisation&empowerment Demographic shift(ageing,migration&identities)Value of knowledge&skills Mental health&well-being Community dynamics&social cohesionSURF TECH TRENDS 2026Artificial Intelligence1.More diversified access to large AI models2.Changing dynamics in Responsible AI3.Increase in co-evolution between hardware and AI4.Collaboration between humans and AI5.From large to small language modelsAuthors Frank Benneker(Universiteit van Amsterdam),Heleen van der Laan (ROC van Amsterdam-Flevoland),Natasha Alechina(Open Universiteit),Rob van de Star(Windesheim),Lars Veefkind(SURF),Simone van Bruggen (SURF),Julius Ceasar Aguma(SURF)SURF TECH TRENDS 2026IntroductionOver the past years,the full realisation of the transformer architecture(a digital neural network that can process vast amounts of data)has completely transformed the Artificial Intelligence(AI)landscape.Most AI research and development has pivoted towards Generative AI(GenAI),a subfield that employs computational models to generate text,images,video,and audio.In many countries,government-funded organisations are developing large AI models for use in fields such as education,research,and healthcare.Unsurprisingly,many organisations are also integrating GenAI into their products and services in pursuit of profitability.Businesses are also seeking real-world applications in the name of efficiency,while also aligning with the responsible deployment of AI in those applications.For example,there is an increase in the use of AI-Concierge,where AI can assist users(like hotel guests)with enquiries around the clock via popular messaging apps.A key adoption bottleneck is the lack of knowledge that the general consumer,as well as many employees in organisations,have about prompt engineering for GenAI.Put simply,for any GenAI model to produce what a user wants,the user needs to know what type of instructions to provide the model with to get a desired result.This has pushed Agentic AI into the spotlight as an obvious solution for the novice AI user.In Agentic AI,the AI system serves the user by operating autonomously to perform pre-defined tasks with little or no human involvement or supervision.Large-scale AI models(which are trained using enormous datasets)are becoming widely available and accessible,and applications built around them are increasingly being integrated into everyday life.The emergence of other new Human-Computer Interactions(HCI)powered by AI and concepts such as AI Love(where humans form intimate connections with AI)is also gaining traction.A major consequence of the advancements in AI is that it is now a significant topic of geopolitical interest and controversy.In essence,AI is a key tool in the political world and as it now represents the greatest asset in the geopolitical game of AI supremacy.Governments are investing more in developing AI models and supporting infrastructure as they seek to establish themselves as AI superpowers to realise the full potential of AI for broader society and national security.The unfortunate result of the race for AI dominance appears to come at the cost of a responsible or ethical AI based on principles such as trustworthiness,explainability,and human-centric AI.ContributorsBas Smit(Erasmus University Rotterdam),Duuk Baten(SURF),Bertine van Deyzen(SURF),Erna Sattler(Leiden University),Damian Podareanu(SURF)SURF TECH TRENDS 2026Artificial IntelligenceMore diversified access to large AI modelsTREND#1Since ChatGPT became available and accessible for widescale use in 2022,large language models(LLMs)have gained massive worldwide traction.For example,ChatGPT reached 1 million users in just a few days after its launch,prompting more diversified access to LLMs and acted as the catalyst for AI to be increasingly part of many daily workflows.Access to AI is facilitated by browser-based tools(ChatGPT,Claude,and DeepSeek)as well as integrations in services and products from vendors like Google(e.g.Docs)and Microsoft(e.g.Copilot,Outlook).Business models range from freemium tiers to pay-per-use APIs.Users now have the option to choose between cloud-based AI solutions or smaller,lightweight models running locally.Concurrently,open-source initiatives have contributed to the varied access to and democratisation of LLMs and tools.This development is reshaping how we write,code,research,and interact with AI,and even each other.MaturityDrivers WATCH PLAN ACTPublic ValuesAutonomyPrivacy|IndependencyJusticeTransparency|Trustworthiness|SustainabilityHumanityWellbeingIndividualisation&empowerment;Mental health&well-being;Value of knowledge&skills;Automation&AI;Connectivity&interactionSURF TECH TRENDS 2026Artificial IntelligenceNumber of GPT usersMost students are using at least one of the available platforms to access a foundation model hepi.ac.uk#library.educause.edu#OpenAI is now the most popular.Other options are Anthropics Claude,Meta,Google options e.g.LM Notebook&Gemini(datastudios.org)#As of late 2024-early 2025,DeepSeek has been getting quite some traction despite security risks(dig.watch)#According to Semrush,OpenAI was the number five website in April 2025()#ChatGPT isnt the only chatbot thats gaining users()#Gartner:why task-specific AI models will take over LLMs()#Rapid rise of Generative AI:the leading companies(iot-)#Growing available models(&providers)on Huggingface(huggingface.co)#Vatican:new Vatican document examines potential and risks of AI(vaticannews.va)#Is the search engine losing ground to AI chatbots?#European AI on the rise openeurollm.eu#swiss-ai.org#gpt-nl.nl#galaxus.nl#swiss-ai.org#SIGNALSSURF TECH TRENDS 2026Artificial IntelligenceIMPACTEducation The diversified access to AI presents both opportunities and challenges for educators by automating mundane tasks,allowing them to focus on teaching and personal guidance,while necessitating adaptation to an AI-driven educational landscape.As students increasingly utilise AI chatbots for learning,productivity and general cognitive offloading,there may be significant implications for traditional tutoring roles and human skills such as critical thinking.Amongst students,the digital divide might widen due to variations in knowledge,access and permission to use.Research In scientific research,AI can accelerate data analysis,enhance text interpretation,and support hypothesis generation.This serves to improve the speed and depth of scientific inquiry.Challenges include concerns over copyright and intellectual property,the risk of disinformation in generated content,and the danger of researchers becoming over-reliant on AI tools.More proposals and papers are produced which increases the reviewing burden.LLMs are increasingly being used as a research partner(research agents/co-scientists).Operations Institutional processes are simplified by automating routine or repetitive tasks,enhancing information management,and enabling smarter workflows.Institutions have to navigate challenges such as potential copyright and data ownership issues,the risk of spreading disinformation,the reduction of human oversight,and concerns regarding the environmental sustainability of large-scale AI implementations.Services and products enabling AI lead to unauthorised data processing/contract breach and DPIAs.SURF TECH TRENDS 2026Artificial IntelligenceChanging dynamics in Responsible AITREND#2The EU is perceived as the global leader in regulation in comparison with the US.The AI Act,the worlds first regulatory framework grounded on risk,transparency,and rights,demonstrates the EU as the defender of Responsible AI.Despite this pioneering legislation,recent geopolitical events mainly in the US-have shifted attention from ethics and human-centric AI towards national security and economic competition.Big tech companies are downsizing their Responsible AI commitments by laying off their ethics boards and prioritising competitiveness over transparency and responsible deployment.However,there are signals that many businesses across the globe do want to implement Responsible AI,though they lack support or mechanisms to act upon it.Although generally small in scale,progress is being made to put Responsible AI programs in place.Nevertheless,the gap between intention,talk,and action on Responsible AI remains.MaturityDrivers WATCH PLAN ACTPublic ValuesAutonomyPrivacyJusticeIntegrity|Equity|InclusionHumanitySafetyGeopolitics&(digital)sovereignity;Concentration of wealth&economic inequality;Compliance®ulationSURF TECH TRENDS 2026Artificial IntelligenceCompanies like Google and others are disbanding their ethics boards and firing ethics-aligned employees #European Commission has withdrawn the proposed AI Liability Act(that would have made using copyrighted material for training models subject to copyright infringement)#Harvard business review(HBR)research:how responsible AI protects the bottom line(hbr.org)#Little more than half(52%)of companies actually have a responsible AI program in place,BCG data shows()#87%of managers acknowledge the importance of responsible AI(RAI)-MIT Technology Review()#Think twice before using DeepSeek:security and trust issues explained (carleton.edu)#The UK has delayed plans to regulate AI as ministers seek to align with the Trump administration()#SIGNALS“There is growing evidence that many big tech companies are backsliding on their commitments to responsible AI.”-Virginia Dignum,Professor AI and Director of AI Policy Lab,Ume UniversitySURF TECH TRENDS 2026Artificial IntelligenceIMPACTEducation Copyright protection for teaching materials is been eroded,and there is a need to educate students in a more effective manner on how to use GenAI responsibly and safely.Students need to learn not to blindly trust GenAI models,and as Virginia Dignum says,educate them in navigating an industry culture that may be increasingly ambivalent towards responsible AI.Research Copyright issues are impacting research involving GenAI models.Researchers require access to independently developed models,rather than just those derived from major tech companies,to avoid legal complications and foster responsible innovation.Industrial collaborations may be affected because of diverging policies on data management and processing.Independent research into AIs impact on society needs to be prioritised.Operations Assessment methods should be reassessed due to a lack of measures from OpenAI to prevent coursework cheating.Clear guidelines for students and staff on using GenAI are needed.Offline models for educational use should be provided to end-users.Caution is required when using GenAI for administrative tasks,especially regarding data protection and ethical issues.SURF TECH TRENDS 2026Artificial IntelligenceIncrease in co-evolution between hardware and AITREND#3While hardware initially facilitates the computing for AI,it is now becoming the limiting factor,forcing AI developments to fit with computing infrastructure.The growing popularity and accessibility of AI models are driving a significant shift in hardware innovations.AI models are increasing in size and complexity faster than the advancements in general-purpose computing technologies for processing.In the previous SURF Tech Trends Report 2023,the surge in specialised AI hardware was highlighted.In addition to this trend,more devices(phones,laptops,etc.)are being produced to support daily AI workflows.Another development driven by this mismatch is the design of AI models to use available computational resources more efficiently.With innovations such as low-precision models and AI-aware hardware implementations,AI is being reshaped,driven by the need to achieve more with less in system capabilities.MaturityDrivers WATCH PLAN ACTPublic ValuesAutonomyIndependencyJusticeEquityHumanityAutomation&AI;Engineering advances&computation;Energy supply&power demand;Raw material scarcitySURF TECH TRENDS 2026Artificial IntelligenceNvidias CES 2025 announcements promise a personal supercomputer ()#The pivot to focusing on AI solutions in most hardware marketing()#The AI revolution will require hardware resources with large energy requirements.This makes enterprise infrastructure a strategic differentiator once again()#Projected growth of AI chip market ()#Many existing companies are pivoting towards making AI hardware #More AI hardware-focused companies coming into the market,e.g,Cerebras #axelera.ai#There is an increase in hardware-aware model design(pytorch.org)#SIGNALS“The gains in AI training performance since MLPerf benchmarks have dramatically outpaced Moores Law.”-David Kanter,executive director of MLCommons()SURF TECH TRENDS 2026Artificial IntelligenceIMPACTEducation Personal AI workstations and increased attention for trusted compute architectures may help mitigate current privacy issues with big tech LLMs,as well as reducing data traffic across networks.High-performance compute for(HPC)for AI will offer students the possibility to handle more complex data tasks,however training may be required.Research More accessible computer hardware with higher performance(personal supercomputers)means that there will be enhanced research capabilities for the research community.A dependency on a handful of vendors for AI-hardware exposes a risk of a potential scarcity of AI-devices as global supply chains come under pressure.Operations Personal AI workstations and more efficient AI models,at lower cost,mean students and researchers are less likely to require large-scale high-performance computing(HPC)systems to experiment with cutting-edge AI.Unlike the computer science,data science,and AI communities who will still need to carry out internationally competitive research.Global demand for resources enabling AI specific chips,could limit hardware availability of edge devices supporting smart campus automation,research and education.SURF TECH TRENDS 2026Artificial IntelligenceCollaboration between humans and AITREND#4Rather than taking over jobs,which is the way that AI is regularly framed,AI is becoming a partner at work and in our daily lives.In certain ways,AI is already taking over routine tasks to allow humans to focus on complex and creative work.Next to routine tasks,AI is already considered a companion to turn to for therapy,identifying ones purpose,and resolving life issues.This AI partnership will be further strengthened by the growth of multi-agent systems.Agentic AI,which has been in development since the late 1990s,promises to significantly enhance human-computer interaction through its natural language interfaces.Additionally,other interfaces are emerging in collaborative robots(cobots)and humanoid robots,enabling more human-like interactions.Furthermore,AI is being integrated as ambient intelligence into various tools and devices.Consider wearables with optics and microphones that can analyse the wearers surroundings and assist in the same manner as a personal assistant or AI companion.A noticeable sentiment about this development relates to the use of LLMs from big tech companies for running the applications.MaturityDrivers WATCH PLAN ACTPublic ValuesAutonomyFreedom of choice|Independency|Privacy|DiversityJusticeInclusionHumanitySafety|Social cohesionIndividualisation&empowerment;Value of knowledge&skills;Automation&AI;Connectivity&interaction;Digital transformation;Concentration of wealth&economic inequalitySURF TECH TRENDS 2026Artificial IntelligenceHealthcareAI assistants like PathAI are helping doctors analyse medical images()#Your next AI wearable will listen to everything all the time()#OpenAIs next big bet wont be a wearable:Report()#Project Astra|Exploring the future of learning with an AI tutor research prototype()#More powerful AI is coming.Academia and industry must oversee it together(doi.org)#Humanoid robotsWill be mass-produced in the US and China over the next three years #Software developmentTools such as Github copilot in Character.AI are set to increase productivity and quality in the software development life cycle #character.ai#SIGNALS“Once AI becomes part of the background,its no longer optional.Its invisible but irreversible.”-Pieter Loman,Utrecht UniversitySURF TECH TRENDS 2026Artificial IntelligenceIMPACTEducation Changes are necessary in the curriculum to meet the new ways of collaborating with AI in professions,and all without compromising public values.Agentic AI may reduce teacher workload by automating assessments,feedback,and lesson planning but raises questions about student agency,bias in feedback,and surveillance concerns.As workplace collaboration between humans and AI has a profound impact on jobs,students will need adaptability and a change in mindset as crucial skills for the future.Students,and teaching staff are becoming a data source for big tech,as the availability of AI tools become more widespread and accessible.Research Researchers could benefit from AI agents to handle repetitive tasks(e.g.,formatting,summarising,and literature scanning).Assuring reproducibility,data privacy,and authorship attribution are more complex for AI agents to handle.By streamlining data analysis,supporting experimental design,and managing laboratory tasks through robotic and predictive systems,AI promises increased automation and efficiency.Ownership of research data and associated questions are exposed with the use of publically available LLMs.Operations AI can increase efficiency and productivity,but care needs to taken with ethical judgements.Importantly,the main goal of education is not to lower costs,but to create a meaningful learning environment for students,as well as a safe and inspiring place to work for educational professionals.Institutions risk vendor lock-in and rising IT costs due to their reliance on embedded AI tools.AI-powered services could optimise scheduling,student support,and administration,lowering operational costs.Human-AI collaborative systems will be fed with huge amounts of physical and personal data,necessitating strict privacy regulations and ethical considerations to safeguard users and promote responsible AI practice.SURF TECH TRENDS 2026Artificial IntelligenceFrom large to small language modelsTREND#5After the revolutionary introduction of LLMs,there is growing interest in Small Language Models(SLMs).These are models with up to 10 billion parameters,in contrast with LLMs,which can have hundreds of billions or even trillions of parameters.Training and using LLMs requires enormous amounts of computational resources.In contrast,SLMs are significantly smaller.Therefore,they are less intensive in terms of data processing,hardware,and training time requirements.SLMs also consume less energy,making them more suitable for applications on smaller devices.SLMs are more accessible to users who want to train and run these models on consumer hardware at the edge of a network,especially for single-purpose devices(e.g.sensors).In addition,SLMs are particularly useful for specific tasks rather than for use as general-purpose tools.MaturityDrivers WATCH PLAN ACTPublic ValuesAutonomyPrivacy|DiversityJusticeTransparency|Inclusion|AccountabilityHumanitySafety|Social cohesionIndividualisation&empowerment;Value of knowledge&skills;Connectivity&interaction;Cybersecurity&trust;Concentration of wealth&economic inequality;Energy supply&power demand;Climate change and global warmingSURF TECH TRENDS 2026Artificial IntelligenceOn-device deploymentGemini Nano on Android:Building with on-device gen AI (io.google)#Qualcomm also enables the running of models like Metas Llama 2 on smartphones using optimised,quantised versions()#Specialised hardwareEuropean tech ecosystems focus on developing edge AI chips designed to deliver server-grade AI power on compact hardware axelera.ai#edge-ai-tech.eu#Privacy-preserving applicationsAIFI pilot started in five Dutch hospitals()#Industry integrationEU-funded research programs(like AI-PRISM)explore embedding AI directly into manufacturing and industrial sectors to facilitate real-time analytics(aiprism.eu)#AI-Powered human-centred Robot Interactions for Smart Manufacturing(cordis.europa.eu)#Social industrial collaborative environments integrating AI,Big Data and Robotics for smart manufacturing (cordis.europa.eu)#Call for more environmentally friendly alternatives(nos.nl)#SLMs are well recognised for their lower energy consumption,providing a low-emission option for institutions looking to minimise their carbon footprint(unesco.org)#Why do researchers care about small language models?(quantamagazine.org)#The transformative potential of AI depends on energy (iea.org)#SIGNALSSURF TECH TRENDS 2026Artificial IntelligenceMore info about AI?Visit surf.nl#IMPACTEducation SLMs and edge AI can enhance personalised learning experiences directly on students devices,ensuring privacy and accessibility.However,institutions must manage infrastructure upgrades and maintain equitable access to avoid technology gaps.New skills in model compression and hardware optimisation will be needed.The affordability of SLM devices and enabling technology may increase the digital divide.Research Researchers can benefit from local,real-time analytics,without extensive computational resources.This facilitates studies in resource-constrained environments.Local devices acting as sensors will be able to process data on the fly,offering greater possibilities for location independent research.Operations Institutions can deploy SLMs and edge AI to enhance operational efficiency,such as automating administrative tasks,while reducing dependency on external cloud services and potentially lowering costs.However,device-level infrastructure and skill development investment are necessary to avoid vendor lock-in and maintain long-term flexibility.Institutions should proactively plan for necessary infrastructure upgrades and training programs to effectively leverage the benefits of edge AI and SLMs.SURF TECH TRENDS 2026Artificial Intelligence1.AR and smart glasses become more prominent2.Democratisation of XR content creation using generative AI3.Increasing AI-enhanced user experience4.Growing big tech competition for the XR stack5.Public investments in XR continue to increaseImmersive TechnologiesAuthors Mark Cole(SURF),Rufus Baas(Media College Amsterdam),Funda Yildirim(University of Twente),Nick van Breda(Avans Hogeschool),Silvia Rossi(CWI),Hizirwan Salim(SURF),Paul Melis(SURF)SURF TECH TRENDS 2026IntroductionImmersive technologies create simulated experiences for users where the boundary between the virtual and physical worlds blurs.Examples of these technologies include virtual reality(VR),augmented reality(AR),and mixed reality(MR).It also encompasses concepts like 360-degree video,spatial audio,and interfaces that provide haptic(touch-based)feedback.Extended reality(XR)is mainly an umbrella term for VR,AR,and MR,and in this chapter,the term immersive technologies is used interchangeably.Immersive technologies have progressed over the last decade due to advancements in computation and AI.This development has led to more mature functionalities from the user perspective(like better user-friendliness and higher levels of experience),and best practices have transitioned to real-world uses(see also SURF Tech Trends 2023 and XR trend update 2024).Organisations across societal domains and industries are recognising where immersive experiences(IX)can add value.Currently,the focus is on technologies tailored to address real-world practical challenges and training individuals for real-life scenarios.The technological focus regarding the development of immersive technologies and their XR applications is shifting.Until recently,the focus was on hardware innovations that led to incremental gains in ergonomics,styling/elegance,miniaturisation,and display quality of VR/AR headsets and smart glasses.Gains in comfort and appearance have emerged as a decisive factor for users to(potentially)adopt these headsets and glasses,followed by affordability,functionality,and content.Currently,the technological developments are moving to middleware infrastructures and platforms supporting the deployment of XR applications.These act as service layers for instructors and operational staff to create or adapt content without specialist support,and it could also include easy-to-use platforms(low coding)to create immersive applications.It is expected that Generative AI(GenAI),virtual assistants,and other AI-based tools will further enhance the functionalities and content capabilities of immersive applications over the coming period.The primary obstacle to general adoption of immersive technologies is not the technological capabilities of XR systems but rather change management within organisations:helping professionals like teachers,researchers,engineers,service technicians,police officers,or military personnel to embrace XR as an everyday tool rather than a lab curiosity or gaming device.ContributorsPablo Cesar(CWI),Remco de Jong(UnboundXR),Michael Barngrover (XR 4 Europe),Keith Mellingen(VRINN),Julie Smithson(METAVRSE&XR Women),Bob Fine(IVRHA),Omar Niamut(TNO),Guo Fremon(Clemson University),Michel Caspers(Simulatie Centrum Maritiem)SURF TECH TRENDS 2026Immersive TechnologiesAR and smart glasses become more prominentTREND#1AR and smart glasses have faced adoption barriers in recent years due to bulky designs.Now,they are lighter,more comfortable,and resemble conventional eyewear.A broader range of products has targeted niche and general markets.Major companies like Apple,Meta,and Google are signalling imminent launches,thus indicating that smart glasses will become mainstream and more widely used.However,it is not expected that smart glasses will fully live up to their potential and become mainstream in the next 15 years.Till then,it is anticipated that AR and smart glasses will be developed mostly for specific use cases.To date,several AR devices and smart glasses have been introduced into the market.While some glasses are still in the beta-testing phase,others are launched softly and in a controlled manner,for example on a country-by-country basis.However,key questions concerning functional added value,appearances,usability,and ethics will remain.MaturityDrivers WATCH PLAN ACTPublic ValuesAutonomyPrivacyJusticeInclusion|Equity|Transparency|HumanitySafety|Meaningful contact|Well-beingIndividualisation&empowerment;Automation&AI;Engineering advances&computation;Connectivity&interaction;Concentration of wealth&economic inequalitySURF TECH TRENDS 2026Immersive TechnologiesSalesMeta Ray-Ban glasses are selling well(2 million already,and production will vastly increase to 10 million per year)()#Promise by big tech companiesGoogle showing off their AR glasses at TED and call them “The next computer”()#Apple Working on smart glasses to beat Ray-Ban Meta,new report claims()#New launches of productsHTC Vive Eagle()#Metas next-gen smart glass()#Product reviews&sneak peek demos()#Amazon developing consumer AR glasses to rival Meta ()#Acceptance perspectives Perspectives on the acceptance and social implications of smart glasses(ris.utwente.nl)#SIGNALS“AR glasses enable seamless assisted reality,allowing for researchers outside a lab to easily view and collaborate on physical lab tasks in real-time.”-Remco de Jong,CEO UnboundXRSURF TECH TRENDS 2026Immersive TechnologiesIMPACTEducation Faster skills transfer is foreseeable in practical lab settings based on for example augmented instructions.Augmented reality functionality of glasses assists in an easier wayfinding and provision of student information through the campus.Easier live caption of education&training activities for students and teachers with impaired vision or language barriers.Attention to the use of AR glasses within the classrooms,exam settings and communal areas,including the processing of biometric data may need a policy review.Research Devices support through augmented reality and forms of user interaction,enhanced remote collaboration with research peers,such as lab team members.Easier accessibility of captured(live)data through smart glasses.Interaction with research information and communication related to open science.Operations Smart glasses in combination with augmented reality and digital overlays support equipment installation,operations,and maintenance.Augmented reality functionality of glasses assists in an easier wayfinding for visitors through the campus.SURF TECH TRENDS 2026Immersive TechnologiesDemocratisation of XR content creation using generative AITREND#2Recently,GenAI and other AI-based tools have emerged to support XR,mainly VR,content creation.This development reshapes and democratises content creation by lowering technical barriers and costs and enabling faster development and deployment.The tools empower actors,such as independent creators,enterprises,educators,and researchers,to proto type ideas and produce 3D-constructed assets and virtual worlds faster,more efficiently,and at scale,without extensive technical skills.In other words,AI is facilitating low-code to no-code XR creations for everyone in the community.The adoption is gradual and use-case dependent.Challenges emerge surrounding copyright,content ownership,skill development in creative design and content creation,and excessive energy consumption.MaturityDrivers WATCH PLAN ACTPublic ValuesAutonomyIndependence|Freedom of choiceJusticeInclusion|TransparencyHumanityPersonal developmentIndividualisation&empowerment;Value of knowledge&skills;Automation&AI;Energy supply&demandSURF TECH TRENDS 2026Immersive TechnologiesPlatform integrationMajor XR platforms are embedding generative AI directly into their toolchains lowering entry barriers:Adobe Project Scenic builds and manipulates full 3D scenes from simple prompts,integrating seamlessly into creative workflows()#Unity 6 includes Generative AI Features for smart NPCs,real-time speech synthesis,and asset generation()#Meta Horizon Worlds allows now to generate entire VR environments via AI prompts,lowering the barrier for social XR world-building()#Faster 3D asset creation with AIGenerative AI tools are significantly reducing time and effort required for 3D asset creation:AI modelling tools can reduce 3D asset creation time by up to 70%for complex scenes(alpha3d.io)#Meta 3D AssetGen generates high-quality meshes with realistic texture and materials in under 30 seconds (assetgen.github.io)#SIGNALS“In the past,users without technical skills werent able to do 3D creation on their own.Now,out of the blue,they have easy to use AI-based tools that help them do something without any 3D modelling experience.”-Keith Mellingen,VRINNSURF TECH TRENDS 2026Immersive TechnologiesIMPACTEducation The AI-XR convergence trend supports faster creation of bespoke XR learning environments once the visual fidelity performance is met by XR platforms and applications for education.Despite the low-code and no-code,teachers might still require assistance and support from experts and skilled professionals to guide students.ResearchThe democratisation of AI-enabled XR content can support rapid prototyping of experimental XR-based setups and access to graphical and 3D-modelled assets for research.Operations Lower production costs for simulations require high levels of functional validation(for example,safety-critical simulations).Facilitating platforms for creation and deployment of XR emerge.SURF TECH TRENDS 2026Immersive TechnologiesIncreasing AI-enhanced user experienceTREND#3AI is enhancing the user experience in real-time by augmenting realism,context-aware interactivity,and personalisation.Predictive models,sensory input(eye tracking,voice recognition),and continuous data analysis enable XR systems to interpret human gaze,gestures,emotions,and spatial mapping in real time.Simultaneously,this adapts the immersive experience to the user without prompting.Virtual assistants enhance engagement,while other AI-based engines optimise graphics performance and provide dynamic content and visual fidelity.Emerging technologies like brain-computer interfaces(BCIs)and neurotechnology could minimise physical efforts when using XR.This AI/XR convergence will lead to AI-enhanced XR that anticipates user needs and behaviours.Maturity WATCH PLAN ACTPublic ValuesAutonomyFreedom of choice|DiversityJusticeInclusionHumanitySafety|Well-being|Personal developmentIndividualisation&empowerment;Community dynamics&social cohesion;Automation&AI;Connectivity&interactionDriversSURF TECH TRENDS 2026Immersive TechnologiesAlways-available,adaptive AI-agents(Ai)Daptive XR platform empowers students and instructors to run fully immersive ()#Digital identity convergence Immersive Technology,blockchain and AI are converging(weforum.org)#AI as storyteller and choice architectEnhancing User Experience in VR Environments through AI-Driven Adaptive UI Design()#Genie 3:Creating dynamic worlds that you can navigate in real-time()#Seamless multimodal engagement User Tracking and Haptic Feedback driving more realistic XR interaction()#Emerging braincomputer interfaces enable hands-free control by translating neural signals directly into commands ()#Meta-review on Brain-Computer Interface(BCI)in the metaverse(doi.org)#Even lower latency gameplay with frame warp()#Meta Quest 4 and Quest 4S()#SIGNALS“Voice recognition,eye tracking these are all about lowering interaction effort for enhanced user experiences.Eventually,with brain-computer interfaces,we could all become XR developers!”-Omar Niamut,TNOSURF TECH TRENDS 2026Immersive TechnologiesIMPACTEducation Conversational AI-based tutors could reduce instructor workload and reach larger student cohorts.AI-driven interactivity(including,for example,gamification)may enhance long-term student engagement,especially amongst younger learners.Conversational agents and personalised feedback systems support student study when human instructors are not available 24/7.Research Context-aware virtual assistants accelerate scientific literature reviews and experiment configuration.Researchers can benefit from intelligent systems with AI-XR functionalities that adapt experimental pathways,offer real-time suggestions,and track engagement metrics for cognitive and emotional responses.OperationsAI chatbots are already being used to answer routine queries from employees/professionals(for example IT or human resources queries),freeing staff time for more complex cases.This could evolve into 3D assistants in virtual or augmented workspaces.SURF TECH TRENDS 2026Immersive TechnologiesGrowing big tech competition for the XR stackTREND#4The XR stack is the set of technologies,components,platforms,and applications that is needed to offer XR/immersive experiences.A key development is the competition between big tech companies(Apple,Meta,Microsoft,and Google)which is a repeating pattern to create complete and leading XR stacks with associated ecosystems.Control over head-mounted display hardware,operating systems,(AI-based)XR engines,middleware,and the adoption of virtual assistants is central to gaining a competitive edge in the XR market.Big tech companies are progressing substantially on their XR stack control as they look at their portfolios of technologies,products,services,and partnerships.On the contrary,these same big tech companies are also participating in standardisation bodies and open XR initiatives.Maturity WATCH PLAN ACTPublic ValuesAutonomyFreedom of choice|Independence|DiversityJusticeEquity|InclusionHumanityWell Being|Social cohesionGlobalisation;Concentration of wealth&economic inequality;Ideologic polarisationDriversSURF TECH TRENDS 2026Immersive TechnologiesMeta opening up their Horizon OS to other manufacturers together with partners,AndroidXR announced with major partnersIntroducing our open mixed reality ecosystem ()#Learn more about Android XR()#Strategic acquisitionsFirst smartphones,now Google acquires another HTC Division and also obtains a non-exclusive license to use HTCs XR intellectual property()#Most manufacturers have announced to use Metas or Googles OS and platform AndroidXR-Samsung,Lynx,Qualcomm,Xreal,Sony:Android XR:A new platform built for headsets and glasses(blog.google)#Meta Horizon OS-Microsoft,Lenovo,Asus Android XR:A new platform built for headsets and glasses(blog.google)#Google demos Android XR glasses at Google I/O 2025 with Gemini integration()#SIGNALS“The most forward-thinking creators are adopting AI tools quickly.Theyre not afraidtheyre using them to push the limits of whats possible in immersive creation.”-Julie Smithson,METAVRSE&XR WomenSURF TECH TRENDS 2026Immersive TechnologiesIMPACTEducationXR stack competition will affect the degree of privacy and data control within XR platforms and applications for educational purposes.Platform lock-in dictates long-term portability and interoperability of(costly)content,and therefore,careful procurement is needed.ResearchAccess to lower levels of software within XR-related platforms and applications for certain research activities will likely be restricted within closed stack systems.This is particularly relevant to consumer-grade price point products available in the market.OperationsThe strategic choice of an XR ecosystem will significantly influence vendor dependence in areas such as interoperability,privacy and control,and ongoing maintenance costs.SURF TECH TRENDS 2026Immersive TechnologiesPublic investments in XR continue to increaseTREND#5Besides the large investments of big tech and other companies in(segments of)the XR stack also governments are starting to recognise XRs potential in innovation,productivity,and economic growth.Therefore,in various domains and sectors,public investments in XR are growing both nationally and internationally,with a shift towards funding more practical applications and use cases.For example,in healthcare(surgical training and mental health therapy),in defence(combat training,drone management and operations),in urban planning,and in education programmes(simulator-trained professionals).In the Netherlands,the establishment of large public-private research and innovation programs indicates substantial national public investment,alongside several Horizon Europe calls focusing on Virtual Worlds.Maturity WATCH PLAN ACTPublic ValuesAutonomyFreedom of choice|Diversity|Independence|PrivacyJusticeInclusion|Transparency|EquityHumanitySafety|Well-beingValue of knowledge&skills;Global trade&tariffs;Digital transformation;Geopolitics&(digital)sovereignty;Compliance®ulation;Critical infrastructureDriversSURF TECH TRENDS 2026Immersive TechnologiesNational investments in education and training in Netherlands npuls.nl#ciiic.nl#dutch.technology#rif-smart.nl#oasis.nl#Defence Five US navy warships get AR tech for remote-assisted repairs()#European Defence Fund(eufundingoverview.be)#Horizon Europe projects XR Projects Financed under H2020 and Horizon Europe(ec.europa.eu)#Varjo suggests a shift to practical applications()#SIGNALS“Healthcare is a big area where investment is set to increase.Theres a lot of funding going towards VR training and applications in the healthcare sector.”-Michael Barngrover,XR4EuropeSURF TECH TRENDS 2026Immersive TechnologiesIMPACTEducationIncreasing public investments in XR accelerate adoption in education,leading to more institutions gaining access to high-fidelity virtual environments like simulators and training applications.ResearchXR focused public funding could shift priorities to applied studies validating measurable quality gains of the real-life usage of virtual environments for training and simulations(such as fewer accidents and higher pass rates).OperationsBetter-organised research and innovation consortia accessing specific public funding pipelines on XR enable faster adoption and shared usage of XR infrastructures.More info about Immersive Technologies?Visit surf.nl#SURF TECH TRENDS 2026Immersive Technologies1.Increased application of FAIR principles to enable digital ecosystems in Europe2.Standardisation of data space architectures for secure and trusted data sharing3.Growing relevance of TRUST principles for data repositories to secure data4.Growing significance of augmented data management5.Emergence of DNA-based data storage to preserve data for a very long timeData ManagementAuthors Mark van de Sanden(GANT,formerly SURF),Jan-Ru Muller(Hogeschool van Amsterdam),Ren van Horik(DANS),Lolke Boonstra(TU Delft)SURF TECH TRENDS 2026IntroductionTodays growth in data is enormous,and this growth is coming from a vast amount of sources such as industrial IoT devices(internet of things),medical imaging systems,synthetic data generators for AI model training,and big science instruments like CERNs Large Hadron Collider(particle physics)and the Vera C.Rubin Observatory(in astronomy).Current scientific experiments can already generate tens of terabytes of data on a daily basis,while future ones will push the scale to hundreds of terabytes per day.While the current data explosion is nothing new,managing these increasing data volumes with current technologies poses challenges which require new approaches and technologies.As an example,storage media like tape and hard disk drives are reaching their physical limits regarding data densities.Simultaneously,the variety of data sets is expanding due to various types of data.Innovative approaches and technologies are necessary,not only for the proper management of vast amounts of data,but also to combine data originating from different domains such as scientific disciplines,industries,and societal domains.This serves to enhance the capability for those managing data to uncover the key insights contained within data sets.Next to the amount of data,the data complexity presents certain challenges.For example,on the multiple roles of organisations like research organisations regarding the processing of large data sets.Not only as a producer of data,not only as an user of data,but also as an actor that combines,enriches,co-creates,and aggregates large SURF TECH TRENDS 2026Data Managementdata sets for and with a variety of other actors.Data management principles and tools help to unlock the value in data.Data management covers the systematic process of handling data throughout its lifecycle:collecting,organising,analysing,sharing,and preserving data while ensuring its integrity,accessibility,and security.AI has shown great promise already in this area,where AI-driven automation can minimise manual effort.Beyond current standard data storage solutions,there is a growing demand for data and content-aware solutions for data management as well as for offering new data insights.Recently,it became clearer how important data management and data preservation are.Looking at developments on data sovereignty,data ownership and security,and open science.These developments are decisive for the way researchers and research organisations cooperate internationally.An example is the recent activity of the research community to preserve large climate data sets stored in the US by saving them on EU-based servers to keep the data freely available for the international climate research community.Besides this data repatriation in the scientific community,national governments and organisations in the EU are also aware of taking stronger measures to secure data ownership.For example,regarding the usage of cloud services by relocating data from the(big tech)servers in the US to servers in Europe.New data management practices and technologies are very much on the horizon and are being formulated to tackle the current and future data challenges.ContributorsTom de Greef(TU Eindhoven),Hans Tonissen(Haskoning),Sara Veldhoen(Beeld en Geluid),Rana Klein(Beeld en Geluid),Damiaan Zwietering(IBM),Laila Fettah(IBM),Matthijs Punter(TNO),Niels Bolding(Health-RI),Lucas van der Meer(ODISSEI),Marie Buesink(Koninklijke nationale bibliotheek),Paolo Manghi(OpenAIRE)SURF TECH TRENDS 2026Data ManagementIncreased application of FAIR principles to enable digital ecosystems in EuropeTREND#1The FAIR principles are guidelines on datasets being findable,accessible,interoperable,and reusable.Increased application of FAIR principles may lead to digital ecosystems,a web of FAIR data and services,where digital resources are not only shared but also semantically linked,automatically interpreted,and reliably reused across domains.Beyond data,a key evolution is the application of FAIR principles to all research outputs,including supporting frameworks such as software,computational workflows,and scientific models.This convergence is facilitated by FAIR Digital Objects(FDOs),which include persistent identifiers and rich metadata,alongside Knowledge Graphs that structure the semantic relationships between these FDOs.The shift to applying FAIR beyond just data only,to include digital research objects,enables automation,reproducibility,and information discovery,while also fostering cross-domain innovation on a larger scale.MaturityDrivers WATCH PLAN ACTPublic ValuesAutonomyJusticeIntegrity I Transparency|SustainabilityHumanityCommunity dynamics&social cohesion;Globalisation;Digital transformationSURF TECH TRENDS 2026Data ManagementInitiatives&organisationsGO FAIR Initiative(go-fair.org)#EOSC(European Open Science Cloud)(eosc.eu)#FDO Forum(fairdo.org)#OpenAIRE(explore.openaire.eu)#ImplementationsAridhia FAIR Data Services-provides researchers with tools for dataset discovery,classification,and metadata browsing()#Fairdata.fi-provides data storage and discovery services to support FAIR principles and ensure long-term preservation(fairdata.fi)#FAIRsharing.org-provides for a curated,informative resource on data and metadata standards,inter-related to databases and data policies(fairsharing.org)#Open Research Knowledge Graph(ORKG)-scholarly communication exploiting the possibilities of digitisation (orkg.org)#Zenodo(zenodo.org)#FAIRimpact project(fair-impact.eu)#Supporting literatureToward the Open Science model:publish your raw diffraction data(pubs.aip.org)#Analysis on open data as a foundation for data-driven research()#Leiden Declaration on FAIR Digital Objects(fdo2022.org)#SIGNALS“The web of FAIR data and services is a keystone vision for the future of digital ecosystems.”-Paolo Manghi,OpenAIRE,CTOSURF TECH TRENDS 2026Data ManagementIMPACTEducation Knowledge Graphs(KGs)are being integrated into Learning Management Systems(LMSs)such as Brightspace and Canvas to model relationships between learning objectives,content,and competencies.Fostering links between learner profiles,learning progress,and course requirements,KGs enable personalised learning through adaptive learning paths.Research The web of FAIR data and services significantly impacts research by accelerating scientific discovery and promoting reusability.FAIR Digital Objects(FDOs),enriched with fine-grained metadata and versioning,enhance the reproducibility,traceability,and reuse of research data,software,and workflows.Operations Knowledge Graphs(KGs)can enhance operational systems by linking data from Learning Management Systems(LMSs)and Student Information Systems(SIS).The integration of KGs enables institutions to map the student journey from pre-admission to graduation and subsequent alumni engagement.SURF TECH TRENDS 2026Data ManagementStandardisation of data space architec-tures for secure and trusted data sharingTREND#2Data spaces allow participants to share,trade,and collaborate on data assets in a manner that is compliant with the participants needs and regulations.These spaces could unleash the enormous potential of data-driven innovation.However,large-scale data sharing is hampered by concerns about trust and the lack of control mechanisms for sharing secure and trusted data.Therefore,standardisation of data space architectures that better support secure and trusted exchanges has received notable attention.The need for structured data spaces has grown,and considering the geopolitical situation,there are significant efforts by the European Commission to stimulate the development of common European Data Spaces.In addition,the maturity heavily varies per domain.The maturity of a metadata standard within the domain enhances the data spaces maturity.Also,the legal basis used within the domain can be very different from domain to domain.MaturityDrivers WATCH PLAN ACTPublic ValuesAutonomyPrivacy|IndependenceJusticeTransparency|Integrity|Accountability|EquityHumanitySafetyCompliance®ulation;Cybersecurity&trust;Value of knowledge&skillsdssc.euSURF TECH TRENDS 2026Data ManagementReportsEuropean Commission update on the status of the common European Data Spaces(digital-strategy.ec.europa.eu)#Critical success factors for Data Space deployment(tno.nl)#Challenges of the clean energy transition and meeting the ambitious targets of the European Green Deal (energy.ec.europa.eu)#European and national level Data Spaces being establishedEuropean Commission strongly supports the development of the common European Data Spaces (digital-strategy.ec.europa.eu)#Future Mobility(marketplace.future-mobility-alliance.org)#European Health Data Space(EHDS)common framework for the use and exchange of electronic health data across the EU(european-health-data-)#SIGNALSReference architectures being developed to support Data SpacesData Space Support Centre publishes version 2 of the Data Spaces Blueprint(dssc.eu)#Draft functional and technical specifications Simpl architecture published(simpl-programme.ec.europa.eu)#Npuls uses the HOSA domain architecture for education and flexibility as a cornerstone(surf.nl)#“In the early days of electricity everything was invented.You didnt come to determine what you could do with it.Now it is standardised,and you can look at where a plug can be connected.This now applies to the standardization of Data Spaces.”-Matthijs Punter,TNO,Data Spaces Support Centretehdas.euSURF TECH TRENDS 2026Data ManagementIMPACTEducationData spaces based on advanced blueprints enable secure,EU-compliant data sharing among educational institutions.They support AI-driven learning,ensure data sovereignty,and reduce dependence on non-European platforms.ResearchData spaces organised by the latest blueprints in research enable secure,faster data access and facilitate cross-border sharing.They enhance scientific collaboration,accelerate research reproducibility,and drive innovation while ensuring ethical and legal data use.OperationsBy enabling secure data exchange amongst members of a data space,operations can be streamlined.The standardised architectures enhance agility,transparency,and alignment with national and European regulations and digital transformation goals.SURF TECH TRENDS 2026Data ManagementGrowing relevance of TRUST principles for data repositories to secure dataTREND#3Data repositories are integral in digital ecosystems,facilitating long-term access to data.Therefore,TRUST principles(Transparency,Responsibility,User focus,Sustainability and Technology)are gaining more relevance recently.These principles ensure repositories are transparent in their operations,handle data responsibly and reliably,use user-focused approaches,use resources sustainably,and deploy technology to secure data management.The TRUST principles complement the FAIR principles and enhance them by making the data infrastructure more reliable and long-term sustainable.In essence,FAIR is about the data,while TRUST is about the repository that manages and preserves the data.So this combination ensures that research data remains a reliable,accessible,and valuable resource for science and society,both now and in the future.MaturityDrivers WATCH PLAN ACTPublic ValuesAutonomyPrivacyJusticeTransparency|Sustainability|Accountability|IntegrityHumanitySafetyValue of knowlegde&skills;Community dynamics&social cohesion;Cybersecurity&trust;Service-oriented&value-based economies;Compliance®ulationSURF TECH TRENDS 2026Data ManagementTransparencyOpen source infrastructure and open governance modelsOpen Science NL drives the transition to open science in the Netherlands,focusing on research data management (nwo.nl)#Open science is central to European research policy,emphasising immediate,unrestricted access to research outputs in shared repositories (research-and-innovation.ec.europa.eu)#User FocusFine-grained data control and consent management (as this is increasingly asked for by the user-community)ODISSEI is a secure environment in the Dutch social science research infrastructure that facilitates easy access,sharing,and processing of sensitive data(odissei-data.nl)#The SIESTA project develops EU-level tools and methodologies for sharing sensitive data,aiming to provide researchers access to confidential information while ensuring privacy and usability(eosc.eu)#ResponsibilityAccountability compliance checksCompliance assessment of the TRUST principles is developed by initiatives related to EOSC(faircore4eosc.eu)#Robust digital preservation strategies and a network of trustworthy repositories ensure data authenticity,integrity,and reliability(EDEN and FIDELIS projects)(eden-fidelis.eu)#SIGNALSSustainabilityManaging human and natural resources responsibly,e.g.by reducing energy consumption(using“green”data centres)The“Green IT maturity model”helps organisations assess their responsibility for the environmental impact of IT and convert intentions into actionable practices(surf.nl)#Digital technology accounts for 5-9%of global electricity use,making energy efficiency vital.The EUs green cloud initiative aims to promote energy-efficient cloud computing (digital-strategy.ec.europa.eu)#TechnologySemantic web standards and APIs enhance interoperability by providing a common language and framework for data exchange,enabling systems to understand and interact with each other more effectively.Researchers,publications,data,funders,etc.are connected by semantic web technologies to form a“Global Open Science Graph”(graph.openaire.eu)#SURF TECH TRENDS 2026Data ManagementIMPACTEducationThe TRUST principles enable data repositories to better facilitate the education sector with the management of data in a more effective manner.They foster collaboration and shared learning through better data sharing and reuse.ResearchResearchers can(re)use and share research data more effectively when data repositories are based on the TRUST framework.OperationsThe TRUST principles can form the basis of the operational activities of next-level data repositories that consider data management dimensions like ethics,user needs,and sustainability.Such an approach will serve to enhance institutional reputation with robust and inclusive data management.SURF TECH TRENDS 2026Data ManagementGrowing significance of augmented data management TREND#4Data volumes are growing rapidly,increasingly generated through automated means,and AI is now accelerating this ongoing trend.Additionally,data sources such as scientific instruments,large sensor networks,and the IoT(Internet of Things)devices in general are becoming more prevalent.Managing these data volumes,extracting information,generating insights,and preserving data value is a challenging task.Augmented data management,a form of AI-based automation,is evolving and radically reducing the manual tasks of data management teams,such as building data orchestration pipelines,assessing data quality,and running repetitive data integration workflows.To properly harness the growing use of AI for data management,data quality is essential.Therefore,AI is recognised as an important pillar for more content-aware and data quality-aware data management solutions.MaturityDrivers WATCH PLAN ACTPublic ValuesAutonomyPrivacyJusticeTransparency|Equity|Accountability|IntegrityHumanityAutomation&AI;Connectivity&interaction;Digital transformationSURF TECH TRENDS 2026Data ManagementAugmented data management initially recognised as trend by Gartner and Deloitte 5 years agoTop 10 Trends in Data and Analytics for 2020()#Tech Trend 2021:Machine data revolution-feeding the machine()#IBM launches new content-aware storage solutions()#Data Direct Networks introduces the Data Intelligence Platform()#Augmented data management becomes matureDevelopment of a maturity model for AI augmented data management(essay.utwente.nl)#Trusted European media data space(TEMS)(beeldengeluid.nl)#AI is transforming research and educationContributing to the web of FAIR data and the uptake of AI (eosc.eu)#How AI is revolutionizing education(weforum.org)#Npuls:Ethical and effective use of AI and data(npuls.nl)#SIGNALSSURF TECH TRENDS 2026Data ManagementIMPACTEducationAI-enabled data management can bring huge advantages to education,developing more personalised learning content,supporting teachers in the assessment process and supporting education by automating administrative tasks.ResearchNext to how AI is transforming how research is being done,augmented data management can automate many repetitive data management tasks in enriching metadata,quality checking,and developing and enabling more content-aware data management solutions for researchers while preserving the value of the data.OperationsAugmented data management only works effectively on quality data.To prevent augmented data management systems from being trained with low-quality data,data managers need to be semantically skilled(on metadata,semantic vocabularies,and semantic thesauri)to ensure that high-quality data is used.SURF TECH TRENDS 2026Data ManagementEmergence of DNA-based data storage to preserve data for a very long timeTREND#5Todays storage solutions(like hard disk drives and tapes)have scaled extensively over the years;however,these types of storage media are reaching the physical limits of data storage densities.A new and promising alternative is using DNA for storing binary data in synthesised strands of DNA.This storage solution which will take years of R&D to be operational-could have a transformational impact on data storage infrastructures,offering potential advantages in data density,encrypted data storage,data durability,long-term data retention,and sustainability.In the meantime,R&D is focused on e.g.workable access speeds the industry needs to develop suitable standards for DNA data storage to facilitate its future deployment.MaturityDrivers WATCH PLAN ACTPublic ValuesAutonomyPrivacyJusticeSustainability|Integrity|AccountabilityHumanityBiotechnology;Automation&AI;Climate change&global warming;Energy supply&demand;Clean water demand;Biodiversity;Raw material scarcityMicrosoft and University of Washington demonstrate first fully automated DNA data storage.Image Microsoft()SURF TECH TRENDS 2026Data ManagementResearch and industry leaders are organising themselvesThe DNA Data Storage Alliance established in October 2020(dnastoragealliance.org)#The DNA Data Storage Alliance joined the Storage Networking Industry Association(SNIA)as a Technology Affiliate in 2022 (snia.org)#First specifications releasedThe DNA Data Storage Alliance released the first two specifications:Sector Zero and Sector One (dnastoragealliance.org)#Data synthesis companies enter the marketTwist Bioscience is a public biotechnology company established in 2013 based that manufactures synthetic DNA and DNA products for customers()#CATALOG has created the worlds first commercially viable device for DNA storage and computation()#Institutes are experimenting with DNA Data StorageResearchers from the University of Washington and Microsoft have demonstrated the first fully automated system to store and retrieve data in manufactured DNA(washington.edu)#Beeld en Geluid stores iconic fragments of EK88 in DNA(nieuws.beeldengeluid.nl)#SIGNALSSound&Vision(Beeld&geluid)stores iconic fragment of European Championship 88 in DNA(nieuws.beeldengeluid.nl)“Data storage capacity of DNA per gram is around 200 million gigabytes,which is millions of times higher than magnetic tape storage densities.”-Tom de Greef,TU/eSURF TECH TRENDS 2026Data ManagementIMPACTEducationAlthough DNA-based data storage has no direct impact on education yet,it could become a new educational topic,or a new way for students to experiment with data storage.ResearchIn research fields where data needs to be preserved for extended periods,or potentially forever due to data importance(for example,National Archive data or endangered languages),future DNA data storage solutions seem to offer long-term possibilities for preservation.Operations DNA data storage could introduce an infrastructural change in how organisations will store,manage,and preserve data in the long term.This would require new skills and industry standards.New companies could emerge specialised in data-oriented DNA synthesis and sequencing.More info about Data Management?Visit surf.nl#SURF TECH TRENDS 2026Data Management1.Trusted digital recognitions are emerging2.More decentralisation of data ownership enabled by SSI solutions3.Emergence of organisational wallets4.Push for transparent supply chains with digital product passports5.Digital Trust Frameworks:from hierarchical to distributed trustDigital TrustAuthors Helmer van Merendonk(Hogeschool Utrecht),Juul de Louw(Koning Willem 1 College),Peter Eikelboom(SURF),Marlies Rikken(SURF)SURF TECH TRENDS 2026IntroductionTrust is the foundation of all digital interactions.It is the confidence that people,systems,and technologies will act reliably,securely,and with integrity.In the digital world,trust means believing that personal data will be protected,systems will work as designed,and technologies will uphold ethical and secure practices.As digital ecosystems become more complex,maintaining and earning that trust becomes not just a technical challenge,but a strategic priority.A shift is taking place in the way that digital identities and the verifiable credentials of individuals,organisations,and objects(physical and digital)are handled.The European Union(EU)is leading this transformation with the Digital Europe Program(DIGITAL).At the heart of this transformation is the European Union Digital Identity(EUDI)Wallet initiative,which seeks to give citizens enhanced and secure control over their digital identities.In an age where misinformation,disinformation,and identity abuse are widespread,storing digital identities and credentials in digital wallets can help improve accountability,safeguard ownership,and reduce the misuse of personal data.By 2027,every EU citizen will have access to an initial version of a digital wallet.This wallet can be used to store and share personal details.This information will relate to both online and offline public and private services across the EU.The EUDI Wallet will be applicable in many aspects of modern life,and it will have implications for how people use and access key identification information.For instance,it will be possible to store a digital version of an individuals driving licence,thus eliminating the need for someone to always have a physical copy on their person.Besides identification information,the EUDI Wallet will provide a seamless recognition of qualifications and provide authorisation across the EU,simplifying education admissions,credentials management,student transfers,job applications,and talent mobility.At the same time,fraud will be reduced,while secure and cryptographically protected identities and credentials will be facilitated.This will help to reduce forgery and increase trust in qualifications.The future of digital trust will give people more control over their own data,by embracing conceptual models such as Self-Sovereign Identity(SSI),along with the use of digital wallets to share their identities and credentials.The success of these initiatives will have a significant impact on citizens as well as public and private organisations.Digital trust assures that the identities and data of people and organisations are handled securely,that digital interactions are reliable,and that their privacy,in terms of their identity and data,is protected.New types of trust networks will create a digital world that is more secure,fair,and trustworthy for everyone.ContributorsYvo Hunink(Founder of Regen Studio),Niels van Dijk(SURF),Peter Leijnse(SURF),Michiel Schok(SURF),Peter Nobels(HU)SURF TECH TRENDS 2026Digital TrustTrusted digital recognitions are emergingTREND#1European frameworks(eIDAS 2.0 and MiCA Regulation)are reshaping the digital landscape with two new concepts:Verifiable Credentials(VCs)and digital assets.Both will enable a shift towards trusted digital recognition of physical and digital objects by 2030.VCs are tamper-proof digital attestations(diplomas or qualifications)issued by trusted institutions or organisations for individuals to store in their digital wallets.Digital assets as tokens,including non-fungible tokens(NFTs),represent digital value,access,or ownership.As it is designed to be transferable:once a token leaves your wallet,its associated rights move too,making them suitable for(value)exchange,single-use recognition,or participation in initiatives.MaturityDrivers WATCH PLAN ACTPublic ValuesAutonomyPersonal empowermentJusticeTransparencey|Equity|Inclusion|IntegrityHumanityIndividualisation&Empowerment;Cybersecurity&Trust;Digital Transformation AspectVerifiable CredentialToken/Digital AssetFunctionReusable proof of achievement or skills,identity,or statusTransferable value,access,or participationReusabilityYes,holder can present it multiple timesNo,it moves with ownershipPrivacyUser-controlled,selective disclosureOften public,depends on token designVerificationVia digital signature and open standards(e.g.W3C,OpenID4VC)Via blockchain consensus and smart contractsSURF TECH TRENDS 2026Digital TrustEuropean regulationeIDAS 2.0 Regulation(eur-lex.europa.eu)#Markets in Crypto-Assets(MiCA)Regulation (eur-lex.europa.eu)#Nederlandse overheid-Toekomstverkenning Digitalisering 2030(rijksoverheid.nl)#Digital assets and tokensAsset tokenization in financial markets:the next generation of value exchange(2025)(reports.weforum.org)#“Trust technologies like verifiable credentials are the new invisible handshake of the digital ageempowering people and organisations to prove who they are and what they know,instantly and securely,at every interaction.”-Adam Eunson,COO,AuvoDigitalEducation,skills&lifelong learningSELFIE for work-based learning(education.ec.europa.eu)#Digital credentials&lifelong learning(unesdoc.unesco.org)#Union of Skills(2024)(commission.europa.eu)#FrameworksVerifiable Credentials&Digital Identity(2024)-Verifiable Credentials Data Model 2.0(w3.org)#OpenID for Verifiable Presentations and Credentials(OpenID4VC)(2024)()#European Digital Identity Wallet Architecture and Reference Framework(2024)(eu-digital-identity-wallet.github.io)#SIGNALSSURF TECH TRENDS 2026Digital TrustIMPACTEducation Learners gain greater control over their learning and career journeys with digital wallets,allowing them to manage and share trusted,verifiable achievements across contexts and borders.This shift enhances mobility,autonomy,and visibility.Educational institutions adopt new roles as issuers and validators within decentralised recognition networks,where formal credentials and purpose-driven tokens co-exist.This transformation supports modular learning pathways,personalised recognition,and broader stakeholder engagement,from employers to societal partners.Research Decentralised reputation models and token-based systems enable more transparent,collaborative ecosystems and new ways of crediting contributions.At a systemic level,eIDAS 2.0 and MiCA provide the legal foundation that enables interoperable,learner-centred ecosystems to emerge and scale.A future-ready education and research landscape,where trust,flexibility,and recognition are embedded,positioning individuals and institutions to thrive in a digitally connected world.OperationsStreamlined processes,reduced administrative tasks,and automated processing of credentials will be realised thanks to interoperable and trustworthy data exchange.SURF TECH TRENDS 2026Digital TrustMore de centra-lisation of data ownership enabled by SSI solutionsTREND#2The increasing value of personal data has led to misuse by many actors in society,ranging from big tech to data brokers and even governments.This has prompted a counter movement focused on user empowerment and data rights protection.In the digital trust and identity domain,this movement toward data decentralisation and related individual control is commonly referred to as Self-Sovereign Identity(SSI).The core principles of SSI include data portability,data minimisation,and access to personal data.Fundamentally,SSI seeks to empower individuals and protect data rights,with the forthcoming EUDI Wallet being a typical SSI solution.MaturityDrivers WATCH PLAN ACTPublic ValuesAutonomyFreedom of choice|Independence|PrivacyJusticeInclusion|Equity|IntegrityHumanitySocial cohesion|Meaningful contact|Well-being|Safety|Personal development|RespectIndividualisation&Empowerment;Cybersecurity&Trust;Digital Transformation;Community Dynamics and Social CohesionSURF TECH TRENDS 2026Digital TrustThe path to Self-Sovereign Identity()#Higher Education and Scientific Research Act(NL)(wetten.overheid.nl)#Data by the Source(digitaleoverheid.nl)#eIDAS 2.0 Regulation-framework for digital identity and authentication (digital-strategy.ec.europa.eu)#A digital ID and personal digital wallet for EU citizens,residents and businesses (ec.europa.eu)#SIGNALSSURF TECH TRENDS 2026Digital TrustIMPACTEducationEducational institutions issue credentials that require data storage for validation.Linking these credentials to individuals necessitates personal information,creating a challenge in designing processes that balance privacy and usability while promoting autonomous decision-making.ResearchThe shift towards Self-Sovereign Identity(SSI)is transforming personal data management.It challenges researchers to adopt good practice and regulate systems that adhere to human rights and legal standards,are technically feasible and promote interdisciplinary research with significant social and policy impact.Operations Overly simple user flows can lead to unconscious decisions and unintended data sharing,while excessive complexity can hinder usability.A balanced approach is essential.Tools based on Self-Sovereign Identity(SSI)principles can empower users to exercise their rights while meeting GDPR requirements.Empowering users reduces institutional control over issued credentials.Institutions must verify identities or wallets before issuing credentials,enabling users to share them across various sectors.SURF TECH TRENDS 2026Digital TrustEmergence of organisational walletsTREND#3Trust and interoperability issues have hampered business interactions across Europe.Announced in 2025,the European Business Wallet(EBW)initiative targets these inefficiencies with a unified,cross-border digital identity solution for organisations.Organisational wallets are digital web-based wallets that provide secure digital identification,streamline data sharing,and facilitate legally valid notifications for companies and other legal entities.Unlike personal wallets,organisational wallets support multi-user access and role-based permissions(such as power of attorney).One-click recognition of organisational identity,status,and attributes can prevent fraud(like fake websites or phishing)and compliance costs,as well as accelerate cross-border collaboration.Organisational wallets must address the complexities of legal entity identity,such as credential lifecycle management,international standards,and integration with business registers.MaturityDrivers WATCH PLAN ACTPublic ValuesAutonomyPrivacyJusticeIntegrityHumanitySafetyIndividualisation&Empowerment;Cybersecurity&Trust;Digital TransformationSURF TECH TRENDS 2026Digital TrustPolicyEU Architecture and Reference Framework(ARF)()#European Business Wallet:digital identity,secure data exchange and legal notifications for simple,digital business (ec.europa.eu)#Use cases and experimentsExploration of EU Digital Identity Wallets for legal entities with Company Passport and iSHARE(coe-dsc.nl)#EU large scale pilots with wallets for businesses (digital-strategy.ec.europa.eu)#SIGNALS“As the European Union moves toward full implementation of the EU Digital Identity Wallet(EUDI),businesses must prepare for a digital transformation in how identities are verified,customers are onboarded,and e-signatures are done.”-Signicat()SURF TECH TRENDS 2026Digital TrustIMPACTEducation Educators and students should benefit from a significant reduction in administrative delays with international exchanges/mobility programs such as Erasmus ,and cross-border teaching or study opportunities.Secure,verified institutional credentials enable quick and trusted confirmation of enrolment,qualifications,and affiliations for internships,further studies,or collaborations.Reduced exposure to phishing and fraudulent communications enhances trust and digital safety in academic correspondence and protects personal and institutional information.Research Reduced risk of impersonation or fraudulent collaboration requests safeguards intellectual property and project research credibility.Verified institutional identities should accelerate the establishment of cross-border collaborations,joint funding applications,and consortium agreements.Operations Instant verification of institutional identity will enable faster execution of information exchange in European public-private partnerships and research collaborations,reducing administrative workload.Role-based digital access and trusted official communications strengthen security by reducing phishing risks,ensuring legal validity,and it also lowers compliance costs through secure,verified information exchange.Integration with internal systems streamlines credential checks and approvals,improving efficiency and optimising processes,although initial investment in staff training and technological infrastructure will be required.SURF TECH TRENDS 2026Digital TrustPush for transparent supply chains with digital product passports TREND#4The EU Ecodesign for Sustainable Products Regulation(ESPR)came into effect in mid-2024.This regulation mandates companies to disclose information regarding the origin of their products and their environmental impact.This mandatory implementation of digital product passports(DPPs)ensures that product data is authentic,reliable,and compliant.Implementing DPPs will be challenging,and production and product data must be verified and accessible via a data carrier.Legislative developments will impact standardisation and DPP service providers.While acts are being finalised,European standardisation bodies are developing the standards for DPPs for adoption by the beginning of 2026.MaturityDrivers WATCH PLAN ACTPublic ValuesAutonomyFreedom of choice|Independence|PrivacyJusticeInclusion|EquityHumanitySocial cohesion|Meaningful contact|Well-being|Safety|Personal development|RespectCompliance and Regulation;Cybersecurity&Trust;Digital TransformationSURF TECH TRENDS 2026Digital TrustEcodesign for Sustainable Products Regulation (commission.europa.eu)#Centre of Excellence for Digital Product Passports (coe-dpp.nl)#Data Sharing Digital Product Passport(tno.nl)#Position paper of Fides(formerly Dutch Blockchain Coalition)about trusted DPPs(munity)#Use cases&pilotsDemonstration of functioning DPPs in different sectors (cirpass2.eu)#Digital product passports:Lessons from an early adopter(British womenswear brand Nobodys Child)()#?SIGNALS“Digital Product Passports will become the main reporting vehicle for all product related compliance information in the European Union.Companies that go beyond compliance and adopt the DPP at the center of their supply chain information ecosystems will see significant additional benefits.”-Yvo Hunink,Founder of Regen Studio and co-writer of the position paper on trusted Digital Product PassportsSURF TECH TRENDS 2026Digital TrustIMPACTEducation Different schools and universities are already discussing the possibilities of using DPPs,the associated challenges with using DPPs,and are working on prototypes.Students are and can research sustainability effects,suitable business models,and the technical(in)capabilities involved with the deployment of DPPs.ResearchDPPs can also be relevant for research and datasets,especially in areas related to sustainability,IT,product lifecycle analysis,and supply chain transparency.Operations Educational institutions can minimise their environmental impact and enhance procurement and campus management.Although not required to comply with the Corporate Sustainability Reporting Directive(CSRD),many institutions are proactively selecting sustainable products and partners.The DPP will improve procurement,increase transparency,and enhance real estate management,enabling campuses to make informed,sustainable choices.SURF TECH TRENDS 2026Digital TrustDigital Trust Frameworks:from hierarchical to distributed trust TREND#5Much of the trust we place in credentials comes from verification by authoritative sources;we cannot rely solely on a persons own claim.This is why educational institutions are considered authoritative when issuing diplomas and credentials.Institutions must verify that the individual attended courses or examinations,and the institution itself must be verifiably accredited.Current trust frameworks that support this system are largely hierarchical.New standards,such as OpenID Federation,enable non-hierarchical trust,complementing existing identity federations and allowing for more independent collaborations.However,this does not eliminate the need for trust.Authorities are still required for accrediting institutions and credentials.Additionally,when processes demand a high level of trust,further technical checks and organisational rules and regulations are necessary.MaturityDrivers WATCH PLAN ACTPublic ValuesAutonomyFreedom of choice|Personal empowerment|Independence|PrivacyJusticeInclusion|Equity|IntegrityHumanitySocial cohesion|Meaningful contact|Well-being|SafetyIndividualisation&Empowerment;Cybersecurity&Trust;Digital TransformationSURF TECH TRENDS 2026Digital TrustOpenID Global StandardsOpenID Federation 1.0 specification()#OpenID API specification for the issuance of verifiable credentials()#Identity federations in practiceOpenID Connect Federation:How to build multilateral federations using OIDC(indico.geant.org)#Identity federations and inter-federation eduGAIN (wiki.geant.org)#EduGAIN OIDC pilot(indico.cern.ch)#SIGNALSSURF TECH TRENDS 2026Digital TrustIMPACTEducationThe role of institutions in the education data ecosystem is evolving,and requests for data and access to tools come from both within and outside of the ecosystem.Therefore,establishing trust in credential authenticity will become more challenging.ResearchThe EU Digital Identity frameworks influence on the adoption of digital wallets in research and education,and its effects on trust frameworks,is subject to ongoing research.OperationsInstitutions must determine how credentials can be reliably attributed to individuals.This requires identity verification at enrolment,during assessments and perhaps even attendance.More info about Digital Trust?Visit surf.nl#SURF TECH TRENDS 2026Digital Trust1.Emerging dual-use of AI in cybersecurity2.More accessible privacy-enhancing technologies to share data3.Rising pressure to prepare for post-quantum cryptography4.Emergence of measures for secure Internet of Things5.The growing need for security protocols to work in the cloudCybersecurityAuthors Anna Gerasymenko(Leiden University),Martine Groen(Hogeschool Utrecht),Mick Deben(MBO Digitaal),Rob Gerritsen(formerly Graafschap college),Nicole van der Meulen(SURF)SURF TECH TRENDS 2026IntroductionThe field of cybersecurity has advanced significantly since computer programmer Bob Thomas created the Creeper virus in 1971.Although this virus,which targeted the Advanced Research Projects Agency Network(ARPANET),was not a malicious virus,it marked an important moment in cybersecurity history.Over 50 years later,cyber attacks have grown increasingly malicious and become a significant threat to society.Technologies,such as Artificial Intelligence(AI)and Internet of Things(IoT),bring new challenges and risks that must be addressed to maintain robust cybersecurity.For example,although AI can help improve defences against cyber attacks,it can also increase the threat by enabling automated malware.While these digital technologies expose organisations to new threats,they are also becoming increasingly essential for organisations to manage and protect themselves.For instance,quantum computing will enable the implementation of complex and potentially very secure cybersecurity protocols.Such technologies offer significant opportunities for both individuals and organisations.Despite these technological advancements,it remains crucial to stay vigilant about potential threats.Understanding the tools and techniques used by cybercriminals and knowing how to secure yourself and your organisation is essential for maintaining(cyber)resilience.In recent years,incidents within the education sector have highlighted the significant damage that can result from a seemingly harmless click by a so-called patient zero.An understanding of how technologies can and could potentially shape the future of cybersecurity is essential.Therefore,this chapter has been included in the SURF Tech Trends to explain how new threats are emerging and to highlight the opportunities arising from technologies that have become integral to daily life.As society grows increasingly reliant on digital systems,cybersecurity and cyber resilience have become vital components of organisational strategy and operations.Organisations must focus on protecting(sensitive)data,complying with evolving policies and regulations(such as the EU Cyber Resilience Act),addressing the vulnerability of IoT devices to malicious attacks or unauthorised control,and mitigating emerging threats.ContributorsZeki Erkin(TU Delft),Jeroen van der Ham-de Vos(University of Twente)SURF TECH TRENDS 2026CybersecurityEmerging dual-use of AI in cybersecurityTREND#1AI is transforming cybersecurity by improving real-time threat detection and efficient responses against these threats.AI models can analyse large datasets and data streams,identify anomalies,and predict attacks,making them a critical tool for security operations.However,cybercriminals are also exploiting AI to improve their attacks.Techniques like AI-driven phishing,deepfakes,and automated vulnerability scanning are making cyberattacks more convincing and scalable.Rising concerns are developments on the deployment of GenAI to generate malware with minimal input and adversarial machine learning.Attackers are manipulating data to deceive AI models such as those used for cyber defence causing the models to overlook threats.MaturityDrivers WATCH PLAN ACTPublic ValuesAutonomyPrivacyJusticeIntegrity|Accountability|TransparencyHumanitySafetyAutomation&AI;Cybersecurity&trust;Engineering advances&computation;Value of knowledge&skills;Weaponisation of knowledge;Digital transformationSURF TECH TRENDS 2026CybersecurityRise of offensive AI capabilitiesCybercriminals are leveraging large language models to craft context-specific phishing content and social engineering scripts Phishing and social engineering in the age of LLMs ()#Back to the hype:an update on how cybercriminals are using GenAI()#With generative AI,social engineering gets more dangerousand harder to spot()#Could cyberattacks turn the lights off in Europe?()#AI-generated deepfakes are increasingly used in identity fraud and disinformation campaigns How deepfakes,disinformation and AI amplify insurance fraud ()#A pro-Russia disinformation campaign is using free AI tools to fuel a content explosion()#The dark side of AI:how deepfakes and disinformation are becoming a billion-dollar business risk()#Open source generative models are enabling low-resource actors to create evasive malware and automate attacks Prediction for open source security in 2025:AI,state actors and supply chains(openssf.org)#OpenAI confirms threat actors use ChatGPT to write malware()#How generative AI is changing how cybercrime gangs work()#“You can see that AI can help both attackers and defenders in cybersecurity.But the main focus remains on getting the basics right,with AI as a tool.”-Jeroen van der Ham-de Vos,University of TwenteSIGNALSSURF TECH TRENDS 2026CybersecurityMitigating adversarial AI risksCyber defenders use adversarial training and robust model tuning to harden AI systems against manipulation and input poisoning.NIST AI risk management framework(nist.gov)#MITRE ATLAS Framework(atlas.mitre.org)#AI red teaming and model auditing are increasingly adopted to identify vulnerabilities in machine learning pipelines before attackers exploit them.Anthropic:challenges in red teaming()#Microsoft Red Team()#Threat intelligence platforms now incorporate adversarial AI detection modules to flag synthetic media,spoofed behaviour,and tampered datasets Recorded Future(recordedfuture)#Darktrace()#TNO and Jungle AI collaborate to detect cyberattack on wind turbine and improve detection capabilities(tno.nl)#SIGNALSDefensive AI integration in cybersecurity operationsAI-powered Security Operations Centres(SOCs)use machine learning to automate threat detection,reduce alert fatigue,and speed up incident response IBM Security,cost of a data breach report()#Microsoft Security Copilot insights()#Behavioural analytics and predictive AI models identify anomalies and pre-empt threats before they escalate into attacks(of any scale)Transforming SOCs with AI:From Reactive to Proactive Security(cloudsecurityalliance.org)#Network&Extended Detection and Response(NXR/XDR)platforms integrate AI to provide real-time attack correlation,autonomous triage,and adaptive defence Microsoft-What is extended detection and response(XDR)?()#Vectra.ai(vectra.ai)#SURF TECH TRENDS 2026CybersecurityIMPACTEducationAI tools offer the potential to secure online learning environments through anomaly detection and behavioural analytics.However,the misuse of AI can undermine educational integrity through automated cheating,deepfakes,or phishing.Securing AI in education requires balancing innovation with ethical and regulatory safeguards.ResearchResearch institutions benefit from AI for advanced threat modelling,data protection,and network monitoring.However,open access policies and collaborative research environments also increase exposure to AI-driven threats.Robust model governance and adversarial resilience
2025-11-03
189页
5星级
Guidance for AI Adoption:Implementation practices Detailed guidance for responsible AI governance v.
2025-10-31
53页
5星级
EricssonMicrowave OutlookOctober Microwave Outlook|October 20252Executive summaryContents02 Executiv.
2025-10-31
12页
5星级
October 2025McKinsey DirectBeyond compute:Infrastructure that powers and cools AI data centers Power.
2025-10-31
11页
5星级
AI trust and governance:Enabling innovation with responsibility October 2025From concept to critical.
2025-10-31
16页
5星级
OECD Digital Government StudiesDigital Government Review of KoreaHarnessing Digital and Data to Tran.
2025-10-30
132页
5星级
Achieving the ASEAN Connectivity Strategic PlanDigital Nations 2025The GSMA is a global organisation.
2025-10-30
42页
5星级
From ideas to impact:a government leaders guide to responsible AI implementationTurning AI ambition .
2025-10-30
31页
5星级
Shaping the Future of AI Talent in Europe:Strategies for Growth and Retention-1-Shaping the Future o.
2025-10-30
81页
5星级
2025 KPMG LLP,a Delaware limited liability partnership and a member firm of the KPMG global organiz.
2025-10-30
20页
5星级
OECD Science,Technology and Innovation Outlook 2025Driving Change in a Shifting LandscapeOECD Scienc.
2025-10-30
251页
5星级
十五五规划建议全文（25页）.pdf
三个皮匠报告：2025银发经济生态：中国与全球实践白皮书（150页）.pdf
2025刘润年度演讲PPT：进化的力量.pdf
三个皮匠报告：2025中国情绪消费市场洞察报告（24页）.pdf
清华大学：2025年AIGC发展研究报告4.0版（152页）.pdf
三个皮匠报告：2025中国AI芯片市场洞察报告（24页）.pdf
毕马威：2025年第四季度中国经济观察报告（82页）.pdf
麦肯锡：2025年人工智能发展态势报告：智能体、创新与转型（英文版）（32页）.pdf
三个皮匠报告：2025中国稀土产业市场洞察报告-从资源到战略武器，中美博弈的稀土战场（25页）.pdf
Gartner：2026年十大战略技术趋势报告（28页）.pdf
0731-84720580
商务合作：really158d
友链申请 (QQ)：1737380874
最新报告
中英对照
全文搜索
报告精选
PDF上传翻译
多格式文档互转
入驻&报告售卖
会员权益
机构报告
券商研报
财报库
专题合集
英文报告
数据图表
会议报告
其他资源
研报
新质生产力
DeepSeek
低空经济
大模型
AI Agent
AI Infra
具身智能
自动驾驶
宠物
银发经济
人形机器人
企业出海
算力
微短剧
薪酬
白皮书
创新药
行业分析
个股研究
年报财报
IPO招股书
会议纪要
宏观策略
政策法规
其他
人工智能
信息科技
互联网
消费经济
汽车交通
电商零售
传媒娱乐
医疗健康
投资金融
能源环境
地产建筑
传统产业
英文报告
其它
行业聚焦
芯片产业
热点概念
全球咨询智库
人工智能
500强
新质生产力
会议峰会
新能源汽车
企业年报
互联网
公司研究
行业综观
消费教育
科技通信
医药健康
人力资源
投资金融
汽车产业
物流地产
电子商务
传统产业
传媒营销
其它
2025年养老经济/银发经济/长寿经济/银发族/老龄化报告合集（共50套打包）
2025年商业航天行业报告合集（共41套打包）
AI、科技与通信
广告、传媒与营销
消费、零售与支付
HR、文化与旅游
金融、保险与投资
能源、环境与工业
医疗制药与大健康
物流、地产与建筑
其他行业
AI ▪ 科技 ▪ 通信
数字化
金融财经
智能制造
电商传媒
地产建筑
医疗医学
能源化工
其他行业
收藏
下载
2025-11-19
AI查数
行业数据
政策法规
商业模式
产业链
竞争格局
市场规模
产业概述
其它
2025年
AI读财报
年报
一季报
半年报
三季报
IPO招股书
社会责任报告
A股
IPO申报
港股
美股&全球
新三板
微信扫码登录
手机快捷登录
账号登录
