保障人工智能的未来：生产中生成人工智能的制衡.pdf

1、Securing the AI Future.The Checks and Balances for Generative AI in P AI in Finance Summit New YorkApril 2024Enterprise adoption of Generative AI How many of you have done a Proof-Of-Concept on a Generative AI use-case within your company?How many of those are actually in production today?Generative

2、 AI-Today64%Face pressureto adopt Generative AI82%Insufficient Visibility&Controls55%Increased RegulatoryLiabilityBig Productivity Boost with Gen AIBIGGER Obstacles in Front of EnterprisesGenerative AI-RisksHallucinationsMislead Users(Airline)Wrong Financial Data(Bank)Incorrect Prescription(Healthca

3、re)JailbreaksChatbots Swearing(DPD)Recommend Competitors(Chevy)Issue Refunds(Ride-Hail App)RegulationsSEC Probe into AI UseEU-AI ActWhite House Executive OrderBreaking down Enterprise concernsWHOis using LLMs?InventoryAuthorizationAccess ControlWHATare the risks?Sensitive Data LeakContent Moderation

4、LLM Attacks&Malicious UsageWHYdo you need to care?CostsComplianceLegal and IP RisksSlow Adoption,Less Productivity1-2 Years to Utilize LLMs in ProductionCurrent processesPilotProduction$MillionsNo ROITimeQuartersLow ProductivityGrowthStagnantNo InnovationEnkrypt AI Framework$Thousands10 x CheaperTim

5、eWeeks10 x FasterGrowthRapid100 x ROIGen AI inProductionPilotReduceRiskIncreaseVisibilityImplementGuardrailsImplementRed-TeamLLM Red-TeamingAutomated and Continuous Red Team TestingKnow your LLMs Vulnerabilities,Choose the Best Model for your ApplicationLlaMa2-7BInjection AttacksLLM Data ProtectionS

6、afe and Compliant Usage of LLMsProtect any Sensitive Information from LeakingPIISensitiveDataPHILLM GuardrailsContext-Aware Guardrails for SecurityPrevent Jailbreaks,Harmful Content,Bias,Malware,HallucinationsLLM VisibilityOperational Transparency and Monitori