1、New defenses,new threats:What AI and Gen AI bring to cybersecurity#GetTheFutureYouWantNew defenses,new threats What AI and Gen AI bring to cybersecurity2Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityTable of contentsExecutive summary04Cybersecurit
2、y to the fore10The AI and Gen AI risk landscape18We rely on AI28Gen AI will reinforce cybersecurity363Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityExploring AI and Gen AI use cases42Recommendations:Using AI and Gen AI to strengthen your cyber def
3、enses50Conclusion 66Research methodology674Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityExecutive summary Cybersecurity incidents on the rise:As the number of cybersecurity incidents rises and the threats including phishing,spear phishing,ransomw
4、are,deepfakes,and fraud schemes grow in sophistication,organizations must enhance their cyber defenses.Our research indicates that 92%of organizations experienced a breach last year,a significant rise from 51%in 2021.The repercussions of these are frequently highly damaging,with around half of organ
5、izations reporting estimated direct and indirect losses in excess of$50 million over the past three years.It is clear that new cybersecurity risks are emerging due to AI and Gen AI.At the same time,the use of these technologies presents an opportunity to enhance an organizations cybersecurity.This r
6、epresents a transformative shift in how security professionals predict,detect,and respond to threats.1.More sophisticated attacks and more adversaries:Threat actors are exploiting AI,including Gen AI,in various ways.Gen AI lowers barriers for these actors,enabling more sophisticated attacks.Typical
7、uses of Gen AI by cybercriminals include phishing,social engineering,deepfakes,malware development,bypassing security controls,exploiting vulnerabilities,automated hacking,creation of malicious GPTs(Generative Pre-trained Transformers),bypassing security controls by mimicking real user behavior.2.Ex
8、pansion of the cyber-attack surface:With 97%of surveyed organizations reporting security incidents related to Gen AI in the past year,organizations must contend with an expanded attack surface.“Prompt injection”attacks manipulate Gen AI models and compromise the integrity of their model outputs.97%o
9、f organizations reported security incidents related to Gen AI in the past year.Three ways in which AI and Gen AI can pose risks:5Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityExecutive summary The external attack surface is becoming increasingly c
10、omplex and multifaceted with the increasing use of AI and Gen AI across various industries.In addition to the traditional attack surfaces that organizations need to protect such as networks,endpoints,data platforms and applications,new vulnerabilities are emerging from these technologies,including c
11、onversational AI agents,apps with AI integration,and multiple AI assistants,advisors and new search tools.Additionally,these technologies can significantly expand the internal attack surface,as internal actors or employees may misuse them such as,by uploading confidential information to external too
12、ls such as ChatGPT.Another concern is shadow AI,where unsanctioned applications are installed and used by employees unaware of company policies,outside of ITs control.3.Lifecycle management of custom Gen AI solutions:The entire lifecycle of Gen AI solutions from enterprise data collection and model
13、customization to development and maintenance must be secured to prevent sensitive data used in customization from being compromised and to ensure the availability and integrity of the solution.Further,Gen AI also brings additional risks,including hallucinations and introduction of vulnerabilities,wh
14、en used for code generation,which can lead to further security issues.Our research finds that organizations are aware of these threats,and about 60%see the need to boost their cybersecurity budgets consequently.Integrating AI and Gen AI into cybersecurity and its benefits:On a positive note,three in
15、 five organizations believe AI to be essential to effective threat response and a majority rely on AI to strengthen their data security,application security,and cloud security.AI enhances threat detection and reporting by providing real-time response capabilities.It significantly reduces analyst fat
16、igue and guides analysts to the most relevant investigation paths,thereby improving both speed and accuracy.Further,organizations also believe Gen AI will strengthen cybersecurity in the long term.The leadership at more than half of the organizations believe Gen AI can advance their security strateg
17、ies.6Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityExecutive summary Exploring AI and Gen AI applications cases for security:AI offers a wide variety of use cases for cybersecurity across IT,OT,and the Internet of Things(IoT),and many organization
18、s are already realizing the benefits.Many are also experimenting with security use cases of Gen AI,such as generating threat intelligence and vulnerability assessments.Enhancing cyber defenses with AI and Gen AI:Organizations must embrace a comprehensive strategy to safeguard their operations.We rec
19、ommend:Develop a clear strategy for integrating AI and Gen AI into existing security systems.Assess the efficiencies gained and risks mitigated relative to the investment into these technologies.Maintain an incident response protocol with actionable instructions for swift,effective action.Continuous
20、ly reassess the security landscape,enabling timely identification of new risks and deployment of adaptive defense mechanisms.Acquire necessary infrastructure,including advanced communication systems,data management solutions,and cloud computing resources.Establish a robust framework,policies,and gov
21、ernance to ensure data safety and integrity,fostering trust in AI models.Focus on model selection and training tailored to organizational needs.Invest in AI and Gen AI-based solutions to integrate with existing security operations centers(SOC)systems enhancing their effectiveness.Gradually integrate
22、 AI agents into cybersecurity operations to assist analysts in responding to incidents and mitigating threats effectively.Ensure ongoing monitoring and updates of AI systems to counter evolving threats.Invest in comprehensive AI cybersecurity training to ensure employees understand AIs and Gen AIs c
23、apabilities and limitations,thereby warranting responsible use.Finally,given the rise in cyberattacks,safeguarding business processes and fostering a culture of risk awareness among employees should be a top priority.7Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring
24、 to cybersecurityWho should read this report and why?Why?In this research,we explore the role of AI and Gen AI in strengthening cybersecurity.We hope our insights will help C-suite executives and cybersecurity leaders to identify use cases of interest and provide recommendations that organizations c
25、an act upon to strengthen their defenses.This report is based on the findings of a comprehensive survey of 1,000 industry executives and in-depth interviews with selected executives.Excluding public-sector organizations,all the organizations surveyed have annual revenue of over$1 billion and 60%have
26、 over$5 billion.All have either already begun to use AI for cybersecurity or are considering it.See the research methodology at the end of the report for more details on the organizations surveyed.Who?This report presents an overview of AI as a key aspect of developing and enhancing cybersecurity re
27、silience,with Gen AI both feeding this drive and benefiting from it in terms of protection for Gen AI projects.This report is written for C-suite executives and cybersecurity leaders working across automotive,consumer products,retail,banking,insurance,telecom,energy and utilities,aerospace and defen
28、se,high-tech,industrial equipment manufacturing,pharma and healthcare,and the public sector.8Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityDefinitionsFor the purposes of this research,we use the following definitions:Artificial intelligence(AI):De
29、velopment of computer systems capable of performing tasks that historically required human intelligence,such as recognizing speech,making decisions,and identifying patterns.AI is an umbrella term that encompasses a wide variety of technologies,including machine learning,deep learning,and natural lan
30、guage processing(NLP).1 Machine learning(ML):This is a subfield of AI that uses algorithms trained on data sets to create self-learning models that are capable of predicting outcomes and classifying information without human intervention.It is used today for a wide range of commercial purposes,inclu
31、ding suggesting products to consumers based on their past purchases,predicting stock market fluctuations,translating text from one language to another,and much more.2 Generative AI(Gen AI):It is a type of AI that has the capability to learn and reapply the properties and patterns of data for a wide
32、range of applications,from creating text,images,and videos in different styles to generating tailored content.It enables machines to perform creative tasks previously thought exclusive to humans.3 9Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurity10C
33、apgemini Research Institute 2023Automotive Supply Chain:Pursuing Long-Term ResilienceCybersecurity to the fore01Julio C.Padilha Chief Information Security Officer at Volkswagen and Audi,South AmericaThe number of attacks weve encountered has doubled over the past four years.”11Capgemini Research Ins
34、titute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityAn unwelcome by-product of the rapid advancement of digital technologies is the exponential increase in security incidents and breaches,causing serious concern for organizations worldwide.A security incident is defined as a
35、ny event that undermines the exposure,integrity,or availability of information systems,potentially leading to significant operational and reputational damage.Corence Klop,Chief Information Security Officer at Rabobank,says:“Over the past year,we had 90 million attacks on our bank.The trend keeps ris
36、ing,which makes technologies like AI crucial.We need to be smarter in our defense strategies as the number of attacks continues to grow.Further,Julio C.Padilha,Volkswagen and Audis Chief Information Security Officer for South America,comments:“The number of attacks weve encountered has doubled over
37、the past four years.”A cyber security expert from a multinational aerospace and defense company elaborates:“We are witnessing a rise both in the number and variety/type of attacks organizations face.Today,threat actors find it increasingly easier to orchestrate attacks,in terms of complexity,frequen
38、cy,and accessibility.”The advent of Gen AI has further complicated this landscape.These technologies,while transformative,have also exposed systems to threat actors.As well as protecting,AI can be weaponized to automate and enhance the sophistication of cyberattacks,making them more difficult to det
39、ect and mitigate.US intelligence officials observe that the government regulations must evolve to keep pace with the recent rapid advancements in AI.4 At the same time,Microsoft-supported studies reveal that 87%of UK organizations are at risk of AI-powered cyberattacks.5 Moreover,Gen AIs ability to
40、generate highly realistic synthetic content poses new risks,such as advanced phishing schemes,misinformation campaigns,and deepfakes.Jason Urso,Chief Technology Officer at Honeywell Connected Enterprise,says:“Prior successful attacks on critical infrastructure involved substantial complexity beyond
41、the capability of an average hacker.However,Gen AI enables less experienced malicious actors to generate malware and initiate sophisticated phishing attacks to gain access to systems and perform automated penetration testing.”6 In our research,we found that:On average,organizations see around 30 sec
42、urity incidents a day.As much as 61%of organizations in the banking sector recorded 1050 incidents a day,the highest among the sectors surveyed.Aerospace and defense(60%),insurance(58%),and telecom(58%)follow closely.Australia,Canada,the Netherlands,and the US have the highest percentages of organiz
43、ations reporting 50-100 incidents daily(22%,21%,20%and 19%,respectively).12Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecuritySource:Capgemini Research Institute,AI and Gen AI in Cybersecurity survey,May 2024,N=1,000 organizations.Figure 1.Around one i
44、n six organizations sees more than 50 incidents a day33%53%14%1%Fewer than 10105050100More than 100Proportion of organizations reporting different volumes of daily cybersecurity incidents13Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityAs figure 2
45、shows,in the past three years,the proportion of organizations experiencing one or more breaches has grown from 51%in 2021 to 92%in 2023.The rapid digitalization of organizations and institutions in response to the COVID-19 pandemic likely contributed to a significant increase in cyberattacks,especia
46、lly in 2022.Threat actors are evolving,with smaller,more agile groups forming to evade law enforcement.Additionally,hackers are increasingly targeting business collaboration tools such as Slack,Microsoft Teams,Microsoft OneDrive,and Google Drive with phishing exploits.7 More interestingly,when we lo
47、oked at all three years,almost all(99%)organizations surveyed have had a breach in one or more of the years.In one massive data breach,a US telecom organization disclosed that malicious actors had stolen the call and text records of more than 100 million consumers from a third-party providers cloud.
48、8 Similarly,customers of an American bank holding company were notified of a potential breach of their data at the beginning of March 2024,including their names,account numbers,and card details.They were urged to monitor their accounts for fraudulent activity over the following 12 to 24 months.9 An
49、American retail firms crowdsourcing delivery service suffered a cyberattack,with malicious actors accessing the sensitive data of some of its drivers between early December 2023 and early February 2024,including social security numbers,drivers license numbers,and other contact information.10 There h
50、as been a rising number of cyberattacks targeting the public sector as well.Incidents of data leaks within the public sector in Singapore rose by 10%in 2023,likely due to the increase in digital services.11 A data breach exposed sensitive information of Canadian government employees.12 The Australia
51、n Government is investigating a“large-scale ransomware”data breach of a health organization,impacting individuals personal and health information.13 99%of organizations surveyed have had a breach at least once in the past three years.Source:Capgemini Research Institute,AI and Gen AI in cybersecurity
52、 survey,May 2024,N=1,000 organizations.Figure 2.Substantial increase in cybersecurity breaches in organizations from 2021 to 2023%of organizations that experienced a cybersecurity breach,20212351%83%92%20212022202314Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring t
53、o cybersecurity Australia experienced a dramatic increase,from 48%in 2021 to 97%in 2023;the US also saw a notable rise,from 61%in 2021 to 95%in 2023.Across sectors,there is a noticeable increase from 2021 to 2023 in the percentage of organizations experiencing cybersecurity incidents.Automotive,for
54、instance,saw an increase from 49%in 2021 to 91%in 2023.In December 2023,a cyberattack on an automotive original equipment manufacturers(OEMs)division in APAC compromised the personal data of 100,000 employees and customers.The breach included names,contact details,and government-issued IDs.In financ
55、ial services,88%and 93%of banking organizations experienced security incidents in 2022 and 2023,respectively,while 89%of the insurance sector reported a breach in both these years.In the past three years,breaches resulted in direct or indirect damages(such as time taken to fix the breach,reputationa
56、l damages,etc.)of more than$50 million for half of organizations.In 2023,another automotive OEMs IT security and data protection policies were compromised when nearly 100 gigabytes of confidential data from customers,employees,and business partners was leaked.The breach could potentially result in a
57、$3.3 billion fine for the organization.14 A US hospitality company suffered a data breach in 2019 that cost over$100 million and exposed the personal information of over 142 million guests.15 One ransomware attack on a multinational in 2023 resulted in data theft that disrupted operations and cost t
58、he company over$27 million in damages.16 Both automotive and insurance sectors have relatively high percentages(12%and 10%,respectively)of organizations experiencing financial damages in excess of$100 million.Source:Capgemini Research Institute,AI and Gen AI in cybersecurity survey,May 2024,N=991 or
59、ganizations that suffered a breach in the past three years.Indirect damages imply time taken to fix the breach,reputational damages,etc.Figure 3.Breaches resulted in average financial damages of$50 million for half of organizations Direct and indirect financial damage to organizations resulting from
60、 a breach in the past three years48%46%5%1%Less than$50mn$50mn$100mn$100mn$500mnMore than$500mn15Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityOrganizations with revenues between$1 billion and$5 billion incur$1.60 in damages for every$100 earned,w
61、hile those with revenues exceeding$20 billion experience$0.40 in damages.Organizations with revenues in between average$0.70 in damages per$100 earned.As organizations across sectors adopt digital technologies and interconnected systems,they also increase their attack surfaces,with an ensuing rise i
62、n security incidents affecting cloud services,machine-speed attacks,17 chatbots,time-sensitive applications,and IoT devices in the past year.Source:Capgemini Research Institute,AI and Gen AI in cybersecurity survey,May 2024,N=1,000 organizations.Figure 4.Cloud saw a nearly 90%rise in security incide
63、nts in the past year 89%rise in cloud related security incidents in the past year.89%10%79%2%19%1%48%5%46%60%5%35%1%56%4%40%1%53%6%41%1%49%5%45%1%40%5%54%1%37%7%56%1%30%5%64%1%CloudMachine-speed attacksChatbotsTime-sensitive applicationsIoT devicesPhishingOT devicesVoice-controlled digital assistant
64、sInsider threatsAdversarial machine learning techniquesZero-day attacksAverage increase/decrease of number of incidents in organization in the past year,by channelDont knowNo changeDecreaseIncrease37%7%56%Corence KlopChief Information Security Officer at Rabobank“The scalability of cyberattacks pose
65、s a significant challenge.However,we can use AI technologies to assist analysts in the Security Operations Center by swiftly searching multiple sources,advising on alert responses,and automating actions to handle high volumes more effectively.”16Capgemini Research Institute 2024New defenses,new thre
66、ats:What AI and Gen AI bring to cybersecurity Cloud services incidents saw the highest reported increase across all sectors,with the highest percentages in aerospace and defense(95%),followed by retail and consumer products(both 92%).Sectors such as aerospace and defense(89%),consumer products(87%),
67、retail and banking(both at 81%),high-tech(78%),and pharma and healthcare(84%),are particularly vulnerable to machine-speed attacks.17Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurity18Capgemini Research Institute 2024New defenses,new threats:What AI
68、and Gen AI bring to cybersecurityThe AI and Gen AI risk landscape0219Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityThe AI risk landscape is evolving rapidly,driven by the latest advancements in technology,increasing adoption across industries,and
69、the emergence of more sophisticated(Gen)AI models.As AI systems become more integral to decision-making,the potential for unintended consequences,biases,and security vulnerabilities grows,necessitating a reevaluation of risk management strategies to ensure trustworthy,secure,and responsible AI deplo
70、yment.The increase in the proportion of organizations that faced breaches(seen in figure 2)and also the increase in machine-speed attacks(as shown in figure 3)could point to the fact that more and more threat actors are relying on AI and Gen AI today to cause cyberattacks.The ways in which AI and Ge
71、n AI can pose risks can be categorized into three areas.1.More sophisticated attacks and more adversaries Threat actors can leverage AI,including Gen AI,in a number of ways.Typical uses of Gen AI by cybercriminals include phishing,social engineering,deepfakes,malware development,bypassing security c
72、ontrols,exploiting vulnerabilities,automated hacking,creation of malicious GPTs(Generative Pre-trained Transformers),bypassing security controls by mimicking real user behavior.For instance,these actors can generate advanced phishing emails,scripts for file manipulation,or code to evade detection.18
73、 In January 2024,the UK governments National Cyber Security Center(NCSC)released an assessment that highlighted that AI will almost certainly increase the volume and heighten the impact of cyberattacks over the next two years.19 Sophisticated threats:Gen AI can lower the barriers for threat actors,r
74、esulting in heightened cyber risks and more sophisticated attacks.Additionally,attackers might manipulate AI systems to produce incorrect predictions or deny customer service.Prompt injection risks:This involves using malicious inputs to manipulate AI and Gen AI models,compromising their integrity.A
75、ttackers can embed harmful scripts and commands in images,causing the model to comply.Multimodal prompt injection attacks can exfiltrate data,redirect queries,spread misinformation,and override safety measures,leading to risks such as fraud and operational sabotage.Social engineering attacks:AI and
76、Gen AI-based social engineering attacks such as deepfakes or phishing emails are particularly challenging to defend due to their high personalization and realism.These attacks use custom lures in chats,videos,or audio,mimicking individuals with remarkable accuracy,targeting multiple systems or indiv
77、iduals with tailored messages.Recently,there have been several notable cases of deepfakes created with Gen AI.20Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecuritySource:Capgemini Research Institute,AI and Gen AI in cybersecurity survey,May 2024,N=1,00
78、0 organizations.Figure 5.More than two in five organizations have suffered financial losses arising from the use of deepfakes An executive at a sports car manufacturer received unexpected messages that appeared to come from the CEO,asking for the signing of several agreements.Using deepfake technolo
79、gy,the attacker conducted a live phone conversation,delivering a voice that closely mimicked the CEOs.However,the attack was thwarted when the executive noticed inconsistencies and realized something was wrong.20 After a design and engineering firm lost$25 mn to a deepfake scam in early 2024,the Asi
80、a branch of another multinational organization became the next victim when a digitally recreated version of its CFO deceived an employee at the Hong Kong office.This fraudulent impersonation led the employee to transfer half a million dollars,purportedly to fund a new branch of the organization.21 A
81、I worms:Further,researchers warn of a new shift in the cyber landscape AI worms.These have the potential to spread from one system to another,potentially compromising data security or deploying malicious software in the process.As Gen AI systems gain autonomy,they are becoming a medium of susceptibi
82、lity to exploitation by malicious actors.These AI worms can exploit this interconnectedness,spreading through Gen AI ecosystems and infecting numerous devices.These worms can potentially compromise critical infrastructure,steal sensitive data,or wreak havoc by disseminating fake news.They can outman
83、euver conventional security by analyzing user behavior to craft personalized attacks.22 Our organization hasencountered deepfake attackswithin the past 12 yearsWeve incurred financial lossesfrom the exploitation ofdeepfake technology45%43%21Capgemini Research Institute 2024New defenses,new threats:W
84、hat AI and Gen AI bring to cybersecurity2.Expansion of the cyber-attack surfaceThe launch of ChatGPT in November 2022 pushed excitement around Gen AI technologies to fever pitch.From 2023,organizations worldwide began to experiment with the new tools,piloting a flurry of use cases.According to our l
85、atest research,nearly one-quarter(24%)of organizations have enabled Gen AI capabilities in some or most of their functions and locations.23 However,as stated previously,the increased adoption of Gen AI brings heightened vulnerability as well as opportunities.In our research,we found that 97%of organ
86、izations encountered breaches or security issues related to the use of Gen AI in the past year.Organizations,today,have to deal with an expanded attack surface area that is becoming increasingly complex and multifaceted.Besides protecting the traditional attack surfaces such as networks,endpoints,da
87、ta platforms and applications,they also need to secure the newer applications enabled by AI and Gen AI such as conversational AI agents,apps with AI integration,and multiple AI assistants,advisors and new search tools.Further,AI and Gen AI can significantly expand the internal attack surface,as inte
88、rnal actors or employees may misuse them.Legal risks:Without stringent governance and oversight,the use of Gen AI can also amplify legal risks,such as exposing trade secrets,proprietary information,and customer data due to inadequate data security measures.Shadow AI:Another concern is the rise of sh
89、adow AI within the organization,where unsanctioned AI applications are installed and used inappropriately.This poses a dual security risk:one from user behavior(such as disclosing confidential information)and the other from the applications themselves(if they have security flaws or vulnerabilities).
90、In a survey,Microsoft found that 75%of knowledge workers around the world use Gen AI at work and that 78%of AI users bring their own AI to work(tools not provided by their organization).24 Our latest research on Gen AI revealed that unauthorized usage among employees is relatively common.Among the 3
91、9%of organizations with a ban or limitation policy,half of them say there is still unauthorized usage of Gen AI in the workplace.25 Furthermore,our recent research on Gen AI for software engineering also reports that 63%of software professionals using Gen AI use it in an unauthorized manner,while on
92、ly 37%use a licensed tool provided by their organization.26 A few organizations have taken the extreme step of completely banning their employees from using AI tools such as ChatGPT.For instance,a multinational organization banned ChatGPT after its engineers accidentally leaked confidential elements
93、 of the companys source code via these tools.27 However,the rise of shadow AI calls to question whether an outright banning of these tools is indeed effective.22Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurity3.Lifecycle management of custom Gen AI
94、solutionsSecuring the entire lifecycle of Gen AI solutions is critical to ensuring the protection of sensitive data and the reliability of the system.From the initial phase of enterprise data collection,where valuable and sensitive information is gathered,to the customization of Gen AI models tailor
95、ed to specific business needs,every stage must be safeguarded.During development and deployment,vulnerabilities,as identified in the insert Mapping and mitigating Gen AI risks,”can arise that could expose confidential data or compromise the systems performance.Implementing robust security measures a
96、cross this lifecycle not only protects sensitive data from being compromised but also guarantees the availability,reliability,and integrity of the solution,allowing organizations to maximize the benefits of AI while minimizing risks.With AI,organizations also have the challenge of ensuring the AI mo
97、dels they build,or use,are free of biases.The adoption of Gen AI could further increase an organizations vulnerability to issues such as hallucination.This is when the model produces an apparently authentic and valid output that it has,in fact,partially or wholly invented.Earlier this year,a Canadia
98、n airline was ordered to pay compensation to a customer after its bot fed inaccurate information to a customer,misleading them into buying a full-price ticket.28 Additionally,organizations are increasingly relying on Gen AI for code generation.While this can improve proficiency,it can also introduce
99、 well-known vulnerabilities(e.g.,the MITRE CWE Top 25 Most Dangerous Software Weaknesses)into the code.29 Two in three organizations are wary of increased exposure to threats While organizations are excited about Gen AIs potential,they are aware of the risks that come with adoption.Most risks associ
100、ated with Gen AI are not novel to application security but rather amplified versions of existing concerns.However,some risks are unique to AI,including model drift,model theft,and data poisoning.Gen AI specifically introduces additional risks and vulnerabilities,such as biased,harmful,or inappropria
101、te content generation,hallucinations,and prompt injection attacks.23Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityIn 2023,Apple restricted the use of ChatGPT and other external AI tools,such as GitHub,for some of its employees.The organization was
102、 concerned that employees could leak confidential data while using these tools.30 Similarly,Amazon has prohibited employees from using third-party Gen AI tools particularly for handling confidential data.This policy is intended to prevent data-ownership issues and protect sensitive company informati
103、on.31 In data poisoning,the AI model is compromised,for example,by injecting malicious data into the training dataset or manipulating the training data to create vulnerabilities.Researchers have,to date,discovered about 100 machine learning(ML)models uploaded to Hugging Face,an open-source platform
104、for ML,that could act as an enabler of the injection of malicious code into user machines.32 Source:Capgemini Research Institute,AI and Gen AI in cybersecurity survey,May 2024,N=1,000 organizations.Figure 6.Two in three organizations are worried about data leakage and data poisoningWe are worried ab
105、out possible leakage of sensitive data/intellectual property theft throughthe training datasets used to train Gen AI modelsData poisoning by malicious actorsis a major concernAdoption of Gen AI has increased riskto organizational securityHarmful breaches caused viamalicious prompts are a major conce
106、rn67%67%64%58%24Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityMost organizations agree they need to increase their security budgets to fortify their defenses Our research into harnessing the potential of Gen AI shows that organizations have achiev
107、ed overall productivity improvements of nearly 8%on average over the past year,and by 2026,they expect significant improvements in operational efficiency,cost reduction,and sales.33 Our current findings also highlight that organizations are increasingly aware of the heightened security risks and the
108、 necessity to enhance their investment in cybersecurity measures.As illustrated in figure 7,most organizations acknowledge the need to increase their allocation toward Gen AI to strengthen their defenses.Source:Capgemini Research Institute,AI and Gen AI in cybersecurity survey,May 2024,N=1,000 organ
109、izations.Figure 7.Nearly 6 in 10 organizations believe they need to increase their security budget to bolster their defenses Organizations and increasingly governments are spending more than ever to protect their databases and critical defense systems from cyberattacks.Cybersecurity now constitutes
110、12%of overall technology budgets,up three percentage points since 2020.34 58%In the wake of Gen AI adoption,we need to massively increaseour security budget toincrease our defenses25Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityMapping and mitigat
111、ing Gen AI risksThe value chain of Gen AI can be complex,with multiple external parties and introduces diverse risks across its stages.A crucial element of developing a secure AI strategy is understanding the various risks the organization is responsible for mitigating.The key to building trust in A
112、I and Gen AI models lies in:Understanding the use case Mapping associated risks Evaluating these risks and Implementing appropriate mitigation strategiesThe below architecture highlights the various risks that can be introduced at each layer.This framework serves as a guide to help organizations map
113、 the risks and create a comprehensive and secure AI strategy.This helps them address the challenges inherent in the Gen AI pipeline,ensuring a robust and trustworthy approach to AI implementation and deployment.26Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to c
114、ybersecuritySource:Capgemini Group Cybersecurity,Trusted AI offer.Figure 8.Gen AI risk reference architectureUser layerApp layerModel layerData layerPromptResponseData leakageData leakageBiased,harmful orinappropriate contentOverrelianceHallucinationPrompt injectionDenial of serviceExcessive permiss
115、ionUnauthorized retrievalSupply chainData leakageSupply chainModel driftModel theftData poisoningData poisoningData leakageData leakageInternal dataLibrariesChat historyTraining dataFine-tuning dataEnd userDeployerBase model/LLMUser/interface27Capgemini Research Institute 2024New defenses,new threat
116、s:What AI and Gen AI bring to cybersecurityEnd users refers to those that interact with Gen AI only through the prompt input and output interface.Deployers refers to anyone who develops,implements,integrates,or manages any part of the Gen AI system.As can be seen in the figure,security concerns or r
117、isks can exist at the data level or at the model layer too,in addition to the application layer and the user layers.In order to secure the initiatives,organizations must understand and minimize each of these risks.Developers are introducing risk-mitigation features.For instance,in Q3 2024 Google is
118、expected to preview its Model Armor protection system,which will enable customers to inspect,route,and protect foundation model prompts and responses,mitigating risks such as prompt injections,jailbreaks,toxic content,and sensitive data leakage.Model Armor will integrate with products across Google
119、Cloud,including Vertex AI.3528Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityWe rely on AI 0329Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityThree in five organizations believe AI is paramount fo
120、r detecting and responding to attacksOrganizations reliance on AI to reinforce their security infrastructure is intensifying,reflecting AIs potentially transformative effect on cybersecurity.Most cybersecurity solutions available on the market and utilized by organizations rely on traditional AI/ML
121、technologies.AI enhances threat detection and response by rapidly analyzing vast amounts of data and identifying patterns and predicting potential breaches.This proactive approach significantly reduces response times and minimizes damage.Moreover,AI-driven automation streamlines routine security tas
122、ks,allowing human experts to focus on more complex issues.In the long term,AIs self-learning should allow the technology to adapt to evolving threats.A cyber security expert from a multinational aerospace and defense company adds:“Currently,we face the challenge of dealing with completely unknown th
123、reats,where traditional methods fail.This is where AI emerges as a crucial tool,as its strength lies in its ability to analyze vast amounts of data and detect specific behaviors.The integration of AI and automation is essential to close the gap between detection and response,ensuring rapid and effec
124、tive cybersecurity processes.”The relevance of AI to cybersecurity is affirmed by 66%of organizations prioritizing its use in this context.Additionally,60%recognize AI as essential to effective responses to cyber threats,emphasizing its strategic significance.60%of organizations recognize AI as esse
125、ntial to effective responses to cyber threats,emphasizing its strategic significance.30Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecuritySource:Capgemini Research Institute,AI and Gen AI in cybersecurity survey,May 2024,N=1,000 organizations.Figure 9.
126、Two in three organizations prioritize the use of AI in cybersecurity As shown in figure 10,data security(76%)and application security(75%)are the areas where organizations most commonly use AI.66%agree that the use of AI in cybersecurity isa high priority for their organization.Use of AI in cybersec
127、urity isa high priority for our organizationWe use AI extensively in our security operationsWithout AI,we would not be ableto identify critical threatsWithout AI,we would not be able torespond effectively to cyberattacks66%65%61%60%31Capgemini Research Institute 2024New defenses,new threats:What AI
128、and Gen AI bring to cybersecuritySource:Capgemini Research Institute,AI and Gen AI in cybersecurity survey,May 2024,N=1,000 organizations.Figure 10.More than three in five organizations use AI in cybersecurity for data security Japan leads in the use of AI in data security(92%),application security(
129、90%),and cloud security(77%).The highest adoption rates in identity and access security are in Japan,the Netherlands,Sweden and Singapore(70%,68%,65%and 64%,respectively).Banking exhibits strong AI adoption in data security(84%),application security and cloud security(both 80%).High-tech has high AI
130、 adoption for application security(88%)and data security(68%),whereas Industrial equipment manufacturing has high adoption in data security(80%)and application security(79%).Use of AI in cybersecurity in the below areas in organizationsDatasecurityApplicationsecurityCloudsecurityNetworksecurityEndpo
131、intsecurityIdentity andaccess securityIoTsecurityOTsecurity70%61%68%59%59%41%76%75%32Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityAI enables a faster response to breaches By automating some elements of threat detection and response,AI minimizes m
132、anual intervention,enhances efficiency,and ensures a cost-effective,rapid and robust defense against evolving cyber threats.“We record billions of cybersecurity events every month.Without AI,it would be impossible to analyze them all effectively,”says Adriano Oliveira,responsible for cybersecurity a
133、t CNP Seguradora,a subsidiary of the CNP Assurances group.Mastercards AI capabilities,for example,enable real-time detection and prevention of payment scams,cutting response times and mitigating financial and reputational losses.36 MetLife uses AI to detect fraudulent claims swiftly,reducing investi
134、gation time and costs by analyzing data patterns,identifying anomalies,and streamlining the fraud detection and resolution process.37 Source:Capgemini Research Institute,AI and Gen AI in cybersecurity survey,May 2024,N=1,000 organizations.Figure 11.More than three in five organizations find AI provi
135、des higher efficiency and accuracy in cybersecurityShare of organizations that have realized various benefits from the use of AI in cybersecurityMore efficientcybersecurity analysisMore accuratebreach detectionImprovedreportingLower cost ofresponse and detectionFaster responseto breachesProactivedef
136、ense strategiesImprovedcompliance64%61%62%59%57%64%64%33Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityOver time,the use of AI by threat actors has rapidly increased,but organizations have not matched this pace in deploying AI for threat detection
137、or remediation(see figure 12).This disparity highlights a substantial opportunity for organizations to advance their AI capabilities to enhance early breach detection and improve response strategies.Source:Capgemini Research Institute,AI in cybersecurity survey,July 2023,N=833 organizations;AI and G
138、en AI in cybersecurity survey,May 2024,N=1,000 organizations.Figure 12.Two in three organizations say they have cut the time taken to detect a security breach by at least 5%Share of organizations that realized time savings after implementing AI in their security operation centers(SOCs)Time taken to
139、detecta breach,2019Time taken to detecta breach,2024Time taken to remediatea breach,2019Time taken to remediatea breach,20246%33%60%21%11%39%29%19%12%40%29%Decrease of 0%-5%More than 10%decreaseNo changeDecrease of 5%-10%13%51%36%34Capgemini Research Institute 2024New defenses,new threats:What AI an
140、d Gen AI bring to cybersecurityMore than 60%of organizations reported that a reduction in their time-to-detect of at least 5%,and nearly 40%of organizations say remediation time fell by 5%or more after implementing AI in their SOCs.Corence Klop from Rabobank comments:“The scalability of cyberattacks
141、 poses a significant challenge.However,we can use AI technologies to assist analysts in the Security Operations Center by swiftly searching multiple sources,advising on alert responses,and automating actions to handle high volumes more effectively.”Adriano Oliveira,Responsible for cybersecurity at C
142、NP Seguradora,a subsidiary of the CNP Assurances group“We record billions of cybersecurity events every month.Without AI,it would be impossible to analyze them all effectively.”35Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurity36Capgemini Research I
143、nstitute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityGen AI will reinforce cybersecurity0437Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityIn the long term,Gen AI will strengthen cybersecurityGen AI will enable advanced th
144、reat simulation and proactive defense strategies.A more than solid 61%of respondents foresee Gen AI strengthening cybersecurity in the long term;a further 62%anticipate it playing a proactive role in vulnerability detection.These insights reflect growing confidence in Gen AIs pre-emptive security me
145、asures.Hlio Cordeiro Mariano,Chief Information Officer at Cooperativa Central Ailos,states:“The interaction and the capacity of Gen AI to learn more about what we are doing,and answer questions based on prompts,could accelerate the identification of issues in the environment.”Gen AIs ability to anti
146、cipate and neutralize sophisticated cyber threats will boost organizational resilience against an ever-evolving digital threat landscape.Frdric Pgaz-Fiornet,Head of Digital Health and Cybersecurity for France,Belgium and Luxembourg at Siemens Healthineers,adds:“We have started using Gen AI for diagn
147、osis and other applications.I am confident that in the coming months or years,AI and generative AI will advance significantly.There are numerous potential applications for generative AI,especially in enhancing proactive measures against attacks.Currently,many CIOs are reactive rather than proactive,
148、lacking full engagement with their monitoring tools during cyberattacks.”Source:Capgemini Research Institute,AI and Gen AI in cybersecurity survey,May 2024,N=1,000 organizations.Figure 13.More than half of organizational leadership believes in Gen AI for securityStatements that apply to organization
149、s regarding the use of Gen AI in cybersecurityOur leadership is a strong advocate of Gen AI to advance cybersecurityOur leadership is not convinced of the potential of Gen AI to advance cybersecurityOur leadership is divided on the potential of Gen AI to advance cybersecurityOur leadership is not su
150、fficiently aware of the potential of Gen AI to advance cybersecurityOur leadership is taking a wait-and-watch approach to Gen AIs use in cybersecurity55%13%3%29%0%38Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurity In both Japan and Australia 62%of o
151、rganizations say their leadership is a strong advocate of Gen AI to advance cybersecurity,the highest among countries surveyed.As many as 70%of organizations in the public sector and 65%of organizations in the high-tech and industrial equipment manufacturing sub-sectors say their leadership is a str
152、ong advocate of Gen AI to advance cybersecurity,the highest across sectors.Over half(54%)say Gen AI can create realistic threat scenarios to enhance cybersecurity analyst training.Frdric Pgaz-Fiornet Head of Digital Health and Cybersecurity for France,Belgium and Luxembourg at Siemens Healthineers“W
153、e have started using Gen AI for diagnosis and other applications.I am confident that in the coming months or years,AI and generative AI will advance significantly.There are numerous potential applications for generative AI,especially in enhancing proactive measures against attacks.Currently,many CIO
154、s are reactive rather than proactive,lacking full engagement with their monitoring tools during cyberattacks.”39Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityMore than half of organizations anticipate faster threat detection and increased accuracy
155、 through the use of Gen AI With the right data and the right model,Gen AIs ability to analyze and interpret vast datasets swiftly allows for early identification of potential threats,though it remains a challenge for more advanced use cases.As a result of its promising capabilities,organizations are
156、 increasingly integrating Gen AI to fortify their defenses,anticipating a marked improvement in their ability to counteract sophisticated cyber threats effectively(see figure 14).Interestingly,fewer than 50%believe Gen AIs ability to save costs.Gen AIs training costs and operating costs are certainl
157、y a cause for concern and can pose a barrier for adoption.38Source:Capgemini Research Institute,AI and Gen AI in cybersecurity survey,May 2024,N=1,000 organizations.Figure 14.More than three in five organizations anticipate faster threat detection by using Gen AI Benefits anticipated from use of Gen
158、 AI in cybersecurityFaster threatdetectionImprovedaccuracyAdvanced threatintelligenceCost savingsProactive threatdetectionBetter regulatorycompliance48%45%46%40%63%54%40Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityNearly three in five organizatio
159、ns believe Gen AI will enhance cybersecurity analysisGen AIs generative capabilities and simulation tools improve overall security measures,making analysts more efficient and effective and substantially freeing them to concentrate on more complex threats.Fifty-seven percent of organizations acknowle
160、dge the importance of specialized training for using Gen AI tools in tasks such as threat detection,incident response,and vulnerability management.Source:Capgemini Research Institute,AI and Gen AI in cybersecurity survey,May 2024,N=1,000 organizations.Figure 15.Over half(58%)of organizations say Gen
161、 AI will empower cybersecurity analysts to concentrate on strategy for combating complex threats58%of organizations say Gen AI will empower cybersecurity analysts to concentrate on strategy for combating complex threatsHow will Gen AI change the roles of cybersecurity professionals?Gen AI will empow
162、er cybersecurity analysts to focus on strategyand building their skillsetsGen AI will enhance cybersecurity analysisGen AI will enable cybersecurity teams to scale their operations and handle increasingly sophisticated cyberattacks58%57%53%41Capgemini Research Institute 2024New defenses,new threats:
163、What AI and Gen AI bring to cybersecurityFrank Hamilton Moraes,IT cybersecurity superintendent,and Luciano Carolino,IT security specialist at Bradesco Bank,comments:“Gen AI can support decision-making,enabling swift action and providing valuable support for security analysts,particularly when utiliz
164、ed in a supervised or semi-supervised manner.This approach is particularly important in complex and critical environments.”Gen AI can also create sophisticated simulations to train security systems and personnel,enhancing preparedness for real-world attacks.JPMorgan Chase uses AI and Gen AI models t
165、o detect fraud by analyzing transaction patterns,identifying anomalies,and improving real-time monitoring,enhancing overall security and fraud prevention.39 Frank Hamilton Moraes IT cybersecurity superintendent at Bradesco Bank“Gen AI can support decision-making,enabling swift action and providing v
166、aluable support for security analysts,particularly when utilized in a supervised or semi-supervised manner.This approach is particularly important in complex and critical environments.”Luciano Carolino IT security specialist at Bradesco Bank42Capgemini Research Institute 2024New defenses,new threats
167、:What AI and Gen AI bring to cybersecurityExploring AI and Gen AI use cases 0543Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityAI use cases across the organizationExciting AI applications in cybersecurity arise across information technology(IT),ope
168、rational technology(OT),and Internet of Things(IoT).Below,we look at a selection of notable use cases.Top use cases:ITOrganizations use AI in IT security to detect threats,automate responses,and analyze vast amounts of data,enhancing cybersecurity for end-point devices,networks,identity,access manag
169、ement,etc.(see figure 16).Source:Capgemini Research Institute,AI and Gen AI in cybersecurity survey,May 2024,N=1,000 organizations.Figure 16.Top five AI use cases in IT AI-driven enforcement of security policies and access controls to protect critical assets and sensitive information.Analysis of dat
170、a from sources including threat feeds and dark web to identify emerging threats and predict attacker tactics,techniques,and procedures(TTPs).Analysis of malware characteristics to predict future infection attempts that signature-based approaches may not detect.AI tools assess the cybersecurity postu
171、re of IT vendors and suppliers,evaluate third-party risk exposure,and ensure compliance with security standards and regulatory requirements to mitigate supply chain vulnerabilities and dependencies.AI can analyze vast amounts of vulnerability data to allow security teams to prioritize critical risk.
172、Vulnerability management and prioritizationVendorrisk managementMalware detectionCyber threat intelligenceSecurity policy enforcement44Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityHere are examples illustrating how organizations are leveraging AI
173、 in IT:Cybersecurity company CheckPoint and Chinas Qihoo 360 Netlab security firm discovered a botnet called“Reaper,”which had affected over 1 million internet-connected devices,including routers and IP(Internet Protocol)cameras.AI threat intelligence can help identify and block botnet traffic,preve
174、nting attacks.40 The US federal government uses an AI platform to analyze billions of events in real time,protecting against all types of attacks,from commodity malware to sophisticated state-sponsored intrusions.41 American Express utilizes AI to analyze customer transactions in real time and ident
175、ify suspicious activity such as unusual spending patterns,location inconsistencies,and known fraudulent activities.42 JPMorgan Chase utilizes AI-powered vulnerability management solutions to monitor extensive networks for vulnerabilities and prioritize patching efforts based on severity and probabil
176、ity of exploitation.43 Cisco uses AI-driven analytics to enhance customers existing identity infrastructure,offering insights into their entire identity population;securing vulnerable accounts;revoking unused and risky privileges;identifying behavioral anomalies;and preventing high-risk access attem
177、pts.44 PayPal uses AI to examine each transaction for red flags and identify and block malicious web content and potential cyber threats.45 45Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityTop use cases:OTOrganizations use AI in OT(hardware and sof
178、tware used to monitor and control devices,processes,and infrastructure in industrial settings)to detect anomalies,predict threats,and automate industrial control,supervisory control,and data acquisition systems(see figure 17).Source:Capgemini Research Institute,AI and Gen AI in cybersecurity survey,
179、May 2024,N=1,000 organizations.Figure 17.Top five AI use cases in OTAnomaly detection in industrial processesAI can detect deviations from normal operating parameters in sensor data to identify potential equipment failure or sabotage attempts in real-time.Vendor riskmanagementAnomaly detectionin ind
180、ustrial processesAI tools assess cybersecurity of OT vendors and suppliers,evaluate third-party risk exposure,and ensure compliance.AI in security systems(e.g.,cameras)can monitor for unauthorized physical access to critical infrastructure.Security policyenforcementAI-driven enforcement of security
181、policy protects critical assets and sensitive information.Digital twinsecurityDigital twin replicas can undergo simulated attacks to test security systems without real-world disruption.Physical securityintegration46Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to
182、 cybersecurityThe below examples illustrate how organizations use AI in OT:The US Transportation Security Administration(TSA)has used facial-recognition technology at various airports nationwide since 2019.The TSAs more than 50,000 agents work across 430 local airports,assessing around 8 million pas
183、sengers each week.46 Honeywell uses AI to swiftly analyze vast amounts of data from industrial control systems,identifying unusual patterns or behaviors.Its AI platform continuously reviews patterns from past incidents and adapts that information to mitigate new emerging threats.47 BBVA,a Spanish mu
184、ltinational financial services organization,has a dedicated AI-driven cybersecurity hub to provide holistic security response to each operational and business element of the bank,including anticipating threats and preparing operational tactics,offering resilience strategies,and protecting BBVAs data
185、 processing centers(DPCs).48 Top use cases:IoTAI enhances IoT security by enabling real-time monitoring of device or plant sensors,threat detection,and automated response(see figure 18).Source:Capgemini Research Institute,AI and Gen AI in cybersecurity survey,May 2024,N=1,000 organizations.Figure 18
186、.Top five AI use cases in IoTAI can develop more sophisticated user authentication methods such as multi-factor and biometric.AI can enforce security policy and access control to protect critical assets and sensitive information.AI analyzes user behavior patterns to improve the robustness of access
187、control.AI tools assess cybersecurity of IoT vendors and suppliers,evaluate third-party risk exposure,and ensure compliance.AI can analyze IoT data streams to identify anomalies in power-consumption patterns,communication,etc.,to indicate potential compromise.Improveduser authenticationEnhancedacces
188、s controlVendorrisk managementSecurity policy enforcementIoT device anomalydetection47Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityAt least two in five organizations have conducted pilot programs using Gen AI for security Gen AI can advance an or
189、ganizations security operations.For instance,LLMs trained and fine-tuned for security use cases,such as the Google Cloud Security AI Workbench,can help analysts identify potential threats.50 Through prompts,analysts can classify,synthesize,and summarize these insights in an intuitive way and preferr
190、ed formats(e.g.,translate attack graphs to human-readable explanations).Further,these tools provide assistance that allows the development of generalist talent to a security analyst role.Organizations have begun integrating Gen AI into their cybersecurity operations.As figure 19 shows,around 40%-50%
191、have initiated proofs of concept(PoC)or pilots.Additionally,nearly three in ten(28%)organizations plan to implement Gen AI in cybersecurity in the near future.Ring and Nest both incorporate AI features and use computer vision algorithms to detect and track motion,send alerts to connected devices,and
192、 offer real-time video streaming via mobile apps.49 28%of organizations plan to implement Gen AI in cybersecurity in the near future.48Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecuritySource:Capgemini Research Institute,AI and Gen AI in cybersecurity
193、 survey,May 2024,N=1,000 organizations,N varies per use case,ranging from 337 to 663.“Implemented”means the use case has been deployed in one or more locations.Figure 19.At least two in five organizations are currently piloting Gen AI for security 4%25%48%23%4%25%48%23%2%25%49%23%3%27%46%23%4%31%44%
194、21%5%33%41%21%5%36%39%20%3%30%46%20%3%27%50%20%5%33%43%19%6%30%45%19%5%29%48%18%4%34%45%17%6%28%50%16%6%27%52%15%3%23%45%29%3%26%42%29%6%21%48%26%3%25%48%25%3%25%47%25%1%26%50%23%Share of organizations implementing Gen AI for cybersecurityImplementedProof of concept/PilotHave plans to implement in n
195、ear futureNot implementedPassword andcredentialsecuritySecure passwordgenerationCyberattacksimulation and trainingPrivacy preservingdata sharingNetworktraffic analysisAutomatedthreat huntingPhishing campaign generationand detectionBehavioralbiometrics andanomaly detectionThreat huntingand forensicAn
196、omalydetectionVulnerabilityassessmentThreatintelligencegenerationDatade-identificationAutomatedthreat responseAutomatedreport generationSecurity policygenerationMalwaredetectionRegulatorycomplianceAdversarialattack detectionDynamic adjustment of IAM policies andtailoring permissionsPhishing attackde
197、tection andprevention49Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityLuciano Valdomiro Dos Santos,Head of Cybersecurity Risk in a Brazilian retail bank,says:“Gen AI can create synthetic datasets and insights that help us test and improve our secur
198、ity measures and controls without exposing real user data.”Below are examples of how organizations use Gen AI in their cybersecurity operations:Symantec,a division of Broadcom Inc.,a US designer,developer,manufacturer,and global supplier of software products,is embedding Gen AI into its security pla
199、tform in a phased rollout.The Gen AI will detect,understand,and remediate sophisticated cyberattacks.51 Mastercard uses Gen AI-based predictive technology to protect future transactions against emerging threats by doubling the detection rate of compromised cards,reducing false positives during card
200、fraud detection by up to 200%,and increasing the speed of identification of at-risk/compromised merchants by 300%.52 Brazilian beauty retail and cosmetics,Grupo Boticrio employs real-time security models to detect,prevent,and respond to potential fraud.53 Per our research,around 20%of organizations
201、have scaled their deployments for their cybersecurity operations,less than the extent of adoption of AI in cybersecurity.The extensive data usage and interactions with models can drive up costs quickly and be a major barrier for large-scale implementation.In addition to the licensing costs of the mo
202、dels(or training costs),significant expenses include maintenance and updates for models and data pipelines.Critical models needing immediate responses,such as those used in live customer service,add further costs for scaling.Managing multiple tools and complex infrastructure also raises operational
203、expenses,complicating large-scale implementations.Secondly,the quality of the Gen AI models output depends on the size and quality of the data on which the models are trained.Without this,organizations may find it difficult to obtain the right results from their Gen AI models.These cost-and data-rel
204、ated concerns could hinder the large-scale adoption of Gen AI for strengthening defenses.50Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityRecommendations:Using AI and Gen AI to strengthen your cyber defenses 0651Capgemini Research Institute 2024New
205、 defenses,new threats:What AI and Gen AI bring to cybersecurityAs AI and Gen AI-based risks continue to rise,organizations security strategies should rely heavily on using the same technologies to defend their assets and operations.Based Source:Capgemini Research Institute analysis.Figure 20.Strengt
206、hening the defenses of your organizations using AI and Gen AIon our survey,interviews,and experience,we propose recommendations for strengthening an organizations defenses below.Develop an AI/Gen AI security strategyRoadmap and selected use casesIn the era of Gen AI,developing a robust security stra
207、tegy is crucial to safeguarding organizational assets and data integrity.Organizations should implement the following:Formulate a clear strategy to integrate AI and Gen AI into existing security systems.Adopt a phased approach that evolves current systems to address nuanced risks and threats targeti
208、ng AI technologies.Both AI and Gen AI have their own applications.For instance,while AI can help with threat or anomaly detection,automating responses to common incidents,and analyzing large datasets,Gen AI can help create realistic phishing simulations,and develop sophisticated scenarios for testin
209、g defenses.Evaluate the efficiencies gained or risks mitigated relative to the investment in the Gen AI tools.Developing a clear strategy for measuring these factors is essential.Beyond the initial investment,it is important to evaluate ongoing operating costs and determine whether the long-term ret
210、urn justifies these expenses.Investing in a tool just because it incorporates Gen AI,without looking into the overall costs,could result in inefficiencies over the long term.Assess your security landscape and risks continuouslyAcquire necessary infrastructureSafeguard business processes and cultivat
211、e a culture of risk awarenessDevelop AI/Gen AI security strategy Establish framework,policies,and guidelinesFormulate plans for integration and monitoringCreate awareness and training programs52Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurity Focus
212、on foundational elements(such as data)and gradually scale up,evaluating associated security risks,as identified in the insert“Mapping and mitigating Gen AI risks.”Identify low-risk,high-value use cases to assess performance and security implications,such as vulnerability patching and enhancing incid
213、ent-response capabilities.Establish a risk reference architecture to standardize and benchmark the identification and mitigation of risks,providing a consistent approach for managing them effectively.Volkswagens Julio C.Padilha comments:“The approach to AI and Generative AI should be grounded in a r
214、isk-based framework,which entails identifying and cataloging all potential risks associated with the technology or project prior to its initiation.”Conduct regular reviews of your cybersecurity strategy and AI policies to adapt to evolving threats and technological advancements.Collaboration among I
215、T teams,legal experts,business leaders,and other stakeholders within and outside the organization,pooling knowledge,experience and resources,is crucial.Luciano Valdomiro Dos Santos from a Brazilian retail bank comments:“Cybersecurity exercises that simulate attacks and prepare for security resilienc
216、e underscore the significance of this cooperation across diverse internal departments and sectors,encompassing not only banks but also various private industries,to ensure continuity in the face of an ever-changing cyber threat landscape.”Incident response protocolsTo effectively manage serious secu
217、rity incidents like data breaches,leaks,ransomware attacks,or loss of sensitive information,organizations should establish a global incident response and management team available 24/7,with strict protocols in place for response and mitigation.The ISO/IEC 27035-1:2016 framework,for instance,provides
218、 protocols for analyzing,assessing,responding to,and containing cybersecurity threats,ensuring alignment with international standards.Outline initial actions to contain and mitigate the incident,such as isolating affected systems,shutting down compromised accounts,and blocking unauthorized access.Le
219、verage Gen AI to provide insights and recommendations to first responders.However,given its relative novelty,it is crucial to keep humans involved in the decision-making process to ensure that actions are taken thoughtfully and appropriately rather than automatically initiating actions from the outs
220、et.Ensure a secure chain of custody for investigation evidence and maintain detailed records of all security incidents.This documentation is essential for future threat analysis,response planning,and proactive vulnerability mitigation.Conduct thorough post-incident analysis to identify gaps in secur
221、ity controls and response procedures to enhance overall cybersecurity resilience.Alexandra Foster,former Managing Director at BT and now an independent consultant,comments:“Its crucial to strengthen your data backup and recovery tools.Many industries collaborate on incident response and prevention,b
222、ut its essential to actively use AI to optimize the backup process and ensure swift recovery in case of data loss.”Integrating AI,especially Gen AI,into incident response planning empowers organizations not only to react swiftly to security incidents but also to anticipate and mitigate future threat
223、s.It can simulate a wide range of cyberattack scenarios,enabling incident response teams to fine-tune their response.53Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityContinuously re-assess security landscape and risksThis pre-emptive approach enabl
224、es the timely identification of new risks and the deployment of adaptive defense mechanisms.Identify critical assets such as sensitive data,IP,and key infrastructure,prioritizing resources and tailoring measures to reduce breach risks and minimize damage.Targeted protection ensures critical operatio
225、ns and information remain secure.Re-evaluating security posture encourages enhanced threat detection and response,helping to maintain system integrity and safeguard sensitive information.This approach also supports compliance with regulatory requirements and fosters a proactive security culture.In o
226、ur research,62%of organizations believed Gen AI will allow them to identify vulnerabilities proactively.Gen AI can further help in interpreting complex regulations and producing detailed reports required for compliance.Julio C.Padilha Chief Information Security Officer at Volkswagen and Audi,South A
227、merica“The approach to AI and Generative AI should be grounded in a risk-based framework,which entails identifying and cataloging all potential risks associated with the technology or project prior to its initiation.”62%of organizations believeGen AI will allow them to identify vulnerabilities proac
228、tively.54Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityAcquire necessary infrastructureAI and Gen AI adoption demands more sophisticated communications,data management,and cloud computing infrastructures,specialized AI processors,and extensive dat
229、a storage.Exploring the synergy between Gen AI and advanced hardware is crucial.Organizations can either develop the necessary infrastructure upgrades in-house or purchase them from specialized providers.Organizations should prioritize hardware security upgrades.High-performance components such as p
230、owerful graphics processing units(GPUs)and specialized AI accelerators enhance Gen AI models,enabling rapid processing and analysis of large volumes of data.This enables AI to swiftly identify anomalies and potential threats in complex networks.Hardware security modules(HSMs)and Trusted Platform Mod
231、ules(TPMs)provide robust encryption and secure-key management.Organizations should also adopt features such as hardware root of trust(RoT,a systemically foundational software component)and fingerprinting to enhance defense layers.54 “The convergence of Gen AI and hardware innovations advances cybers
232、ecurity by enabling faster,more accurate threat detection,enhancing data protection through secure computing environments,and improving user-authentication processes,”affirms Mohit Sagar,Chief Executive Officer and Editor-in-Chief at OpenGov Asia,a content platform,initiating dialogue across public-
233、sector CIOs and technology experts.55 As AI expands in data centers and at the edge,integrating AI-based security mechanisms into data centers and network infrastructure will be crucial.However,faster hardware and processors also consume more energy and contribute to carbon footprints.Instead of buy
234、ing GPU farms and HSMs,organizations should consider leveraging the cloud to expand the required capabilities.Utilizing cloud resources can improve sustainability by optimizing the use of shared infrastructure,which leads to more efficient resource utilization and a reduced carbon footprint.55Capgem
235、ini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityEstablish framework,policies,and guidelinesData pipelines and readinessA robust,well-integrated,and scalable data platform ensures data safety and integrity,which are vital to nurturing trust in AI models.Fu
236、rther,the effectiveness of AI,specifically Gen AI in cybersecurity depends on the size and quality of the data,and the algorithms used to analyze it.Currently,neither the data volume and quality nor the algorithms are sufficiently advanced for the widespread use of Gen AI in cybersecurity aside from
237、 a few targeted applications where data and algorithms are reliable.To address this,organizations should:Identify and classify data sources,files,and unstructured data,especially confidential data(e.g.,customer information,business transactions data,etc.).Track and evaluate data sources for accuracy
238、,completeness,consistency,timeliness,and reliability.Record frequency and purpose of data access to identify dependencies and potential bottlenecks and prevent breaches.Moreover,such data platforms must scale seamlessly as organizations grow their workforces,build complex data infrastructures,and ma
239、nage larger volumes of data.Corence Klop from Rabobank affirms:“I would prioritize organizing your data,ensuring you have a centralized repository to work from.This begins with establishing a single,comprehensive source of data.Rather than starting from scratch,use existing resources and conduct exp
240、eriments to see what works for you.”Governance policiesAI and Gen AI raise important questions regarding data governance,intellectual property(IP),bias mitigation,and responsible utilization of AI-generated content.Establishing a dedicated team or department to oversee AI governance at the organizat
241、ional level is crucial.Highlighting the significance of this governance mechanism,Hlio Cordeiro Mariano from Cooperativa Central Ailos explains:“Were establishing an Innovation team to engage all departments effectively.Our focus spans beyond security leads,encompassing how the company navigates exp
242、eriments,tests,and simulations.Were crafting a process to prioritize impactful actions and demonstrate their value.There are numerous market alternatives,not all are tailored to our specific needs.Our approach involves starting small,testing for value generation,and scaling promising initiatives swi
243、ftly.Governance,planning,and aligning expectations are crucial discussions,particularly when integrating AI to benefit our business comprehensively.”Further,such teams within organizations need to:Clearly articulate and document policies governing the use,storage,and transmission of data in AI and G
244、en AI systems.Establish policies and guidelines for the development and deployment of AI and Gen AI tools to ensure ethical practices and effective governance.Define clear guidelines for employees on the use of AI and Gen AI tools.Emphasize adherence to data privacy regulations and highlight the rep
245、ercussions of misuse.This empowers employees to make informed decisions and reduces organizational risk.The CISO at a payment solutions company in Brazil comments:“Anyone can potentially compromise the LLM,which is the primary security risk we are concerned about.Additionally,privacy breaches are al
246、so a significant concern;sensi-tive information should not be accessible to just anyone.Even within a closed environment,access to all company information should be restricted.These are among the challenges we currently face.”Control and restrict data access to necessary stakeholders through role-ba
247、sed access control and techniques such as blocking,hashing,and limiting platform connectivity to external networks.56Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurity Facilitate regular reassessment of all models in use and their generated outputs to
248、 refine and modify frameworks.The CTO at a multinational clothing company comments:“Organizations must ensure that their models adhere to privacy principles and regulatory requirements.Another critical issue is ethical considerations and bias.Biases present in training data can lead to discriminator
249、y outcomes.Additionally,AI models may produce artifacts that need to be understood and managed.Therefore,ensuring the quality and diversity of outputs related to gender and other factors is essential for the practical application of generative AI.”Be transparent about the privacy policy and give the
250、 customer easy access to resources explaining the underlying logic of AI algorithms,and offering clarity on your methodology to identify,eliminate,and prevent bias.Consider establishing specialized roles within the security team to oversee these efforts.“Certain organizations are appointing a Chief
251、trust officer within their security functions,which exemplifies the convergence where evolving regulations are blurring traditional roles and responsibilities.Now,cybersecurity frameworks are interlinking with considerations of AI trustworthiness,encompassing both risks and opportunities,”said Alexa
252、ndra Foster,former Managing Director at BT.CDOs should work closely with the CISOs and CIOs in ensuring the data integrity and data quality.By designating roles such as these,organizations can effectively leverage technology and data in fighting cyberattacks.Review your vendors policies regarding da
253、ta handling,storage,deletion timelines,and model training.Look for details on traceability,log history,anonymization,and other essential features.Solicit feedback from diverse stakeholders,including technology experts,business professionals and users,to evaluate the potential impacts and implication
254、s of AI applications.There is a clear need for increased collaboration between government and the private sector to manage complex technology platforms.Comply with security and notification requirements under latest regulations such as European Unions AI Act,Network and Information Systems(NIS)direc
255、tive,and GDPR.57Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityAlexandra Foster Former Managing Director at BT and now an independent consultant“Certain organizations are appointing a Chief trust officer within their security functions,which exempl
256、ifies the convergence where evolving regulations are blurring traditional roles and responsibilities.Now,cybersecurity frameworks are interlinking with considerations of AI trustworthiness,encompassing both risks and opportunities”58Capgemini Research Institute 2024New defenses,new threats:What AI a
257、nd Gen AI bring to cybersecurityModel selection and trainingTo reduce the carbon footprint of LLMs,organizations should limit training.Using reputed and reliable LLMs specifically on raw cybersecurity logs minimizes false positives,reducing unnecessary alerts and clarifying genuine threats.They prov
258、ide robust attack simulations and explore what-if scenarios,which are crucial to testing existing alerts and defenses.56 To gain more flexibility with LLMs,organizations can explore open-source models such as Metas Llama and fine-tune them to meet their specific needs.Instead of solely focusing on L
259、LMs,organizations can also consider using small language models(SLMs)that have fewer parameters,require less data and training time,are targeted for specific use cases,and have smaller carbon footprints.SLMs also have smaller attack surfaces,making them less susceptible to adversarial attacks.57 The
260、 Brazilian payment companys CISO comments:One of the main challenges we face with AI in security is the high rate of false positives and negatives.This makes it difficult to trust AI in production.We need to ensure accuracy before deploying AI solutions on a large scale.In the future,organizations w
261、ill likely use a combination of LLMs and SLMs to meet their cybersecurity needs.Therefore,they should also focus on developing a custom Gen AI deployment pipeline to manage these models effectively.Organizations can also explore custom GPT models,which are similar to tailored cyber tools.These model
262、s learn the organizations specific language,adapt to its nuances,maintain constant vigilance to detect anomalies,and learn continuously.Organizations must train such models on relevant data such as logs,incident reports,and threat intelligence.Organizations should synchronize them with the existing
263、tech stack,enabling seamless communication with firewalls,intrusion detection systems and other security protocols.It should be noted that these custom GPTs are also susceptible to vulnerabilities.Setting up guardrails against misuse and,further,segregation of development and production environments
264、 will help in mitigating risks.Maintaining a clear boundary between testing and production systems and closely monitoring and validating model performance and integrity is essential.When training models,organizations do need to note that models trained on sensitive data should also be considered as
265、sensitive.Hence,it is also crucial to safeguard these models with the same level of security and confidentiality as that of the data they are trained on,to prevent unauthorized access.To safeguard these models and the model weights from unauthorized access and theft,organizations should:Establish a
266、security plan that includes centralizing weights on a limited number of access-controlled systems and restricting access.Ensure interfaces for accessing the model are hardened against data-exfiltration attempts.Integrate confidential computing techniques to secure model weights during processing and
267、 minimize attack surfaces.58 Further,organizations can follow the federated learning(FL)approach which is a decentralized ML approach where training occurs across multiple devices,sending only model updates instead of raw data.This method enhances privacy by keeping personal information on local dev
268、ices,improving models collaboratively without centralizing sensitive data.This approach,therefore,supports threat detection,anomaly identification,malware detection,and predictive analysis without sacrificing confidentiality of data.59 Additionally,as the softwarization of chips gain prominence,orga
269、nizations attack surface increases.This expanded attack surface can lead to serious security issues,such as unauthorized access to sensitive data,manipulation of chip functions,or even full system control by attackers.Therefore,introducing robust security measures at the chip level is equally crucia
270、l.59Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityFormulate plans for integration and monitoringIntegrate with existing SOC solutionsGen AI revamps SecOps by enhancing the current SOC capabilities,aiding in automation,data interpretation,suggestin
271、g best practices.As threats grow more sophisticated,organizations must pivot towards AI-driven solutions for enhanced detection and faster response.Organizations should invest in AI-based solutions that can autonomously identify threats and block them.By crafting specific response playbooks,and opti
272、mizing workflows,AI also empowers security teams to efficiently prioritize,detect,and remediate issues.Today,organizations could leverage a number of solutions available in the market to detect threats such as deepfakes and avoid bias in AI system outputs.Similarly,setting up guardrails against comm
273、on attacks will help in ensuring that the AI systems do not turn rogue.It is essential to dismantle silos and foster cross-platform collaboration and promote a cohesive and unified security strategy.This integration enables comprehensive monitoring and protection of all facets of an organizations di
274、gital infrastructure.60 Furthermore,to safeguard against prompt injection risks and the autonomous nature of AI systems,its crucial to implement an additional security layer that continuously monitors and intercepts potentially rogue commands.This secondary system should be designed to wrap around t
275、he primary AI system,acting as a gatekeeper that can scrutinize and filter commands before they reach the core AI functions.Adopting a zero-trust approach where every input is treated as potentially compromised and scrutinized can help mitigate these risks and enhance overall security.Deploy AI agen
276、tsOrganizations should strategically integrate AI agents into their cybersecurity operations.They are designed to function independently,plan,reflect,pursue higher-level goals,and execute complex workflows with minimal or limited direct human oversight.61 In cybersecurity,such agents operate autonom
277、ously and monitor network traffic,detect anomalies,respond to threats in real time,and actively search for threats without human intervention.Additionally,AI agents simulate attacks,identify vulnerabilities,and develop defense strategies.This iterative process involves a collaborative ongoing effort
278、 among these agents.As well as ecosystemic collaboration,organizations must also train employees to collaborate with these advanced systems.As the volume of the attacks increases,AI agents that operate within certain thresholds are critical in defending an organizations operations.However,these agen
279、ts do require safeguards.Our latest research on Gen AI shows that 57%acknowledge the need for robust control mechanisms before integrating AI agents into their operations,and 73%insist that humans must verify and,if needed,intervene in AI decisions.62 A careful balance is required between utilizing
280、such autonomous agents and maintaining oversight due to the risks they present.60Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityContinuous monitoringOrganizations must maintain continuous monitoring and updates of AI and Gen AI systems to defend ag
281、ainst evolving threats.They should:Monitor,measure,audit,and log metrics to ensure the responsible,ethical,and secure deployment of AI models.Implementing scoring mechanisms provides real-time insights into the risk level associated with each input and output,helping users make informed decisions an
282、d maintain robust oversight.Monitoring for model drift over time and recalibrating the models will ensure the output is reliable.Focus testing on AI-specific risks,including jailbreaks,prompt injection,and issues related to coherence,readability,and toxicity of the generated content.Additionally,ass
283、ess for biases and conduct red teaming exercises to strengthen the models security posture.Invest in real-time behavioral pattern matching.As social engineering attacks increase,organizations must be able to detect and prevent any behavior that is out of the norm for an employee.Invest in existing s
284、olutions and tools to detect Gen AI signatures and patterns such as deepfakes,and proactively prevent potential attacks and mitigate threats effectively.Create awareness and training programsIn our research,58%of organizations mentioned a shortage of talented cybersecurity professionals.Additionally
285、,63%acknowledge the difficulty in integrating Gen AI into their existing security solutions due to talent limitations.Consequently,over half(51%)of organizations today are investing in comprehensive AI cybersecurity training programs.These programs foster a deeper understanding of AI capabilities,li
286、mitations,and ethical considerations,ensuring responsible usage.Emphasizing the significance of diverse training programs,Alexandra Foster,former Managing Director at BT,states:“In many organizations today,there is a strong emphasis on basic security awareness and training,particularly in areas like
287、 phishing.I believe theres great potential to expand these efforts to include comprehensive programs on social engineering and malware.This could involve incorporating simulations and leveraging gamification for effective training.Moreover,these initiatives should extend beyond just the cybersecurit
288、y team to encompass all departments,fostering a security-first culture that ensures everyone understands and values their role in maintaining security.”61Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityEnhanced awareness and training lead to improve
289、d threat detection,response strategies,and overall cybersecurity posture.Frdric Pgaz-Fiornet from Siemens Healthineers says:We have hackers within our company who continuously test our systems.We conduct internal testing and also engage the local Computer Emergency Response Team with hacking experti
290、se to ensure our system hardening.User awareness and education in AI cybersecurity ensure that individuals recognize potential threats and understand how to respond.This reduces human error,strengthens overall security,and promotes a culture of vigilance,enhancing the effectiveness of AI-driven defe
291、nses.Upskilling programs in AI and Gen AI for cybersecurity bridge the gap between cybersecurity and AI.Highlighting the importance of Gen AI training,a CISO from an automotive company adds:“Before gaining access to Gen AI,employees must complete mandatory training to understand how to responsibly h
292、andle and utilize the platform.This ensures that sensitive information isnt inadvertently shared.Its crucial to educate employees on what type of information is appropriate for input into the system or platform before they begin using it.”Further,highlighting the importance of educating employees on
293、 Gen AI,Frederic Jesupret,Group Information Security Officer at Allianz Partners,states:“Its essential to have checks in place to ensure that generative AI doesnt enter a negative or erroneous loop,as has been observed in some cases.This underscores the need to elevate the capabilities of my employe
294、es.While generative AI can handle a greater volume of events than manual processes,human oversight remains critical to verify conclusions periodically and ensure ethical considerations are upheld.“Over half of the organizations(56%)in our research believe that Gen AI will significantly redefine the
295、roles and responsibilities of cybersecurity professionals within the next 23 years.58%of organizations mentioned a shortage of talented cybersecurity professionals.63%of organizations acknowledge the difficulty in integrating Gen AI into their existing security solutions due to talent limitations.51
296、%of organizations today are investing in comprehensive AI cybersecurity training programs.62Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurity56%of organizations believe that Gen AI will significantly redefine the roles and responsibilities of cyberse
297、curity professionals within the next 23 years.Awareness programs should extend beyond end-users and security teams to include data scientists and engineers as well,who play a crucial role in the security of the models they develop and customize.Additionally,its important to evaluate the practices of
298、 all relevant personassuch as Gen AI users,developers,data scientists,and cyber and infrastructure teamsto ensure they align with security policies and best practices.63Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityHlio Cordeiro Mariano Chief Info
299、rmation Officer atCooperative Central Ailos“Were establishing an Innovation team to engage all departments effectively.Our focus spans beyond security leads,encompassing how the company navigates experiments,tests,and simulations.Were crafting a process to prioritize impactful actions and demonstrat
300、e their value.There are numerous market alternatives,not all are tailored to our specific needs.Our approach involves starting small,testing for value generation,and scaling promising initiatives swiftly.Governance,planning,and aligning expectations are crucial discussions,particularly when integrat
301、ing AI to benefit our business comprehensively.”64Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecuritySafeguard business processes and cultivate a culture of risk awareness With the rise of social engineering attacks,it is essential for organizations to
302、 embed security awareness into the organizational mindset.Employees must be trained to recognize and report potential threats swiftly.By cultivating a culture of risk awareness,organizations can ensure that employees remain vigilant and proactive in safeguarding against security breaches.A heightene
303、d sense of awareness and critical thinking skills(such as analyzing the context and corroborating from trusted sources)from employees can help organizations counterbalance the threats posed by malicious actors.Additionally,cyberattacks often highlight the issues in business processes.Organizational
304、hierarchies rely on trust in the individuals,with instructions often conveyed via email or workflow tools.With the increased use of digital communication and the ability of threat actors to intercept these communications,its crucial to verify the authenticity of these interactions.AI can address thi
305、s by incorporating real-time risk assessments into business processes.For instance,when executing high-risk transactionslike transferring moneyAI can score the activity,assess risk,and enable informed decision-making.Leveraging AI to evaluate risky transactions ensures greater security and protects
306、people and processes besides assets.Ultimately,human oversight over autonomous AI systems and a clear demarcation of duties can help ensure efficiencies while remaining secure.In the trifecta of people,process,and technology,it is essential that all these three elements seamlessly integrate to creat
307、e a resilient and robust defense.65Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityFrederic Jesupret Group Information Security Officer at Allianz Partners“Its essential to have checks in place to ensure that generative AI doesnt enter a negative or
308、 erroneous loop,as has been observed in some cases.This underscores the need to elevate the capabilities of my employees.While generative AI can handle a greater volume of events than manual processes,human oversight remains critical to verify conclusions periodically and ensure ethical consideratio
309、ns are upheld.66Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityAI and Gen AI are increasingly being exploited by malicious actors,who are leveraging these technologies to enhance their attacks.The threat landscape is continuing to become more sophi
310、sticated,with threat actors leveraging AI and Gen AI to run attacks at scale,posing significant risks to cybersecurity.With the volume of threats increasing more than ever,organizations must turn to these same technologies to mount a strong defense against cyberattacks.Organizations will need an inn
311、ovative edge in the fight to remain secure against malicious actors.They are increasingly relying on AI for fast and accurate detection and reporting of real-time threats and,nuanced anomalies,enabling proactive defense strategies.While the adoption of Gen AI exposes organizations to some new threat
312、s both from outside actors and from employees,Conclusionit also provides opportunities to enhance security.At the same time,they must also understand the risks that are associated with Gen AI adoption and take necessary mitigative actions.To harness AI and Gen AI effectively,organizations should emb
313、race a culture of continuous re-assessment of the security landscape,building the infrastructure and establishing adequate framework and guidelines,as well as establishing robust employee training and awareness programs.They should develop new monitoring and control mechanisms and integrate them int
314、o their existing response processes.This will allow organizations to tap into the full potential of these technologies,creating a resilient cybersecurity posture that will be pivotal for safeguarding their most valuable assets and nurturing trust along the value chain.67Capgemini Research Institute
315、2024New defenses,new threats:What AI and Gen AI bring to cybersecurityResearch methodologyWe conducted a targeted survey of 1,000 organizations that have either considered AI for cybersecurity or are already using it,across 12 sectors and 13 countries in AsiaPacific,Europe,and North America.They hav
316、e annual revenues of$1 billion and over.We carried out the global survey in May 2024.We provide the distribution of these respondents and their organizations below.Organizations by headquarter locationOrganizations by sectorUnited StatesUnited KingdomFranceGermanyCanadaJapanBrazilNetherlandsItalySwe
317、denAustraliaSpainSingapore15%9%9%8%8%6%6%6%6%6%6%5%10%AutomotiveConsumer productsRetailBankingInsuranceIndustrial equipment manufacturingAerospace and defensePharma and healthcareHigh techPublic sectorEnergy and utilitiesTelecom9%9%8%8%8%8%8%8%8%8%8%10%Source:Capgemini Research Institute,AI and Gen
318、AI in cybersecurity survey,May 2024,N=1,000 organizations.68Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityOrganizations by annual revenuePublic sector organizations by annual budgetsRespondents by functionSource:Capgemini Research Institute,AI and
319、 Gen AI in cybersecurity survey,May 2024,N=1,000 organizations.57%40%19%17%$1 billionlessthan$5 billion$5 billionlessthan$10 billionMore than$20 billion$10 billionlessthan$20 billion28%15%26%8%$50mn$100mn$100mn$500mn$5bn$10bn7%More than$10 billion$1bn$5bn18%$500mn$1bnInformation securitySecurity ope
320、rations centerSecurity operationsRisk management(security risks)25%25%24%27%69Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityTo supplement the survey findings,we also conducted in-depth discussions with 18 executives from organizations using AI and
321、/or Gen AI in their cybersecurity defenses.The study findings reflect the views of the respondents to our online questionnaire for this research and are intended to provide directional guidance.Please contact one of the Capgemini experts listed at the end of the report to discuss specific implicatio
322、ns.70Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurityReferences13.ABC News,“Australian government investigating large-scale ransomware data breach of script provider MediSecure,”May 2024.14.Ekran,“7 examples of real-life data breaches caused by insi
323、der threats,”February 2024.15.The Record,MGM Resorts says cyberattack cost$100 million,October 2023.16.Security Week,Johnson Controls ransomware attack:data theft confirmed,cost exceeds$27 million,February 2024.17.Machine-speed attacks are cyberattacks that occur at a pace that surpasses human capab
324、ilities,often leveraging automation,AI and ML to carry out the attack.The goal is to overwhelm defenses and exploit systems before human operators have a chance to react.18.Microsoft,“Staying ahead of threat actors in the age of AI,“February 14,2024.19.National Cyber Security Centre,“The near-term i
325、mpact of AI on the cyber threat,”January 24,2024.20.Fortune,“Ferrari exec foils deepfake attempt by asking the scammer a question only CEO Benedetto Vigna could answer,”July 2024.21.SCMP,“Hong Kong employee tricked into paying out HK$4 million after video call with deepfake CFO of UK multinational f
326、irm,”May 2024.22.Cybersecurity Asean,“AI Worms are crawling up as new AI parasites invade your devices,”May 2024.23.Capgemini Research Institute,“Harnessing the value of generative AI 2nd edition:Top use cases across sectors,”July 2024.24.Microsoft,“2024 Work Trend Index Annual Report,”May 8,2024.25
327、.Capgemini Research Institute,“Harnessing the value of generative AI 2nd edition:Top use cases across sectors,”July 2024.26.Capgemini Research Institute,“Turbocharging software with gen AI,”July 2024.27.Forbes,Samsung bans ChatGPT among employees after sensitive code leak,May 2023.28.The Guardian,“A
328、ir Canada ordered to pay customer who was misled by airlines chatbot,”February 2024.29.Negri-Ribalta C,Geraud-Stewart R,Sergeeva A,Lenzini G.A systematic literature review on the impact of AI models on the security of code generation.Front Big Data.2024 May 13;7:1386720.doi:10.3389/fdata.2024.138672
329、0.PMID:38803522;PMCID:PMC11128619.1.Coursera,accessed in August 2024.2.Ibid.3.Capgemini Research Institute,“Harnessing the value of generative AI:Top use cases across industries”,July 2023.4.Forbes,“Gen AI and its malicious impact on the cyber-physical threat landscape,”April 2024.5.bid.6.VentureBea
330、t,Gen AI is the power surge cybersecurity vendors need to reduce the risks of losing the AI war,December 2023.7.Check Point,“Check Point research reports a 38%increase in 2022 global cyberattacks,”January 2023.8.Verdict,“AT&T in the crosshairs after a massive breach of customer data,”July 22,2024.9.
331、Electric,“High-profile company data breaches,”June 2024.10.Cybernews,“Attackers penetrate Walmarts Spark driver portal,”February 2024.11.The Straits Times,“Incidents of data leaks in Spore public sector up 10%,with 201 cases recorded in 2023,”July 2024.12.CS Hub,“IOTW:Data breach exposes sensitive i
332、nformation of Canadian Government employees,”November 2023.71Capgemini Research Institute 2024New defenses,new threats:What AI and Gen AI bring to cybersecurity30.The Wall Street Journal,“Apple restricts employee use of ChatGPT,joining other companies wary of leaks,”May 2023.31.Times of India,“Amazo
333、n has a warning for employees using AI at work,”February 2024.32.Dark Reading,“Hugging Face AI platform riddled with 100 malicious code-execution models,”February 2024.33.Capgemini Research Institute,“Harnessing the value of generative AI 2nd edition:Top use cases across sectors,”July 2024.34.IANS research security budget benchmark report,data as of October 3,2023.35.Google Cloud,“Advancing the ar