2834 - 强化 AI 管道：使用 IBM Guardium 实现端到端安全.pdf

1、Orlando,FLOctober 69IBM TechXchange 20252834Fortifying AI Pipelines:End-to-End Security with IBM GuardiumDevan Shah,STSM,Chief Architect Data Security,IBMMatt Simons,Program Director Product Management,IBMFortifying AI Pipelines84%64%64%Enterprises are embracing but have concernsSource:“Enterprise g

2、enerative AI:State of the market”2023 IBM Institute of Business Value Study Of enterprises face significant pressure to accelerate generative AI initiativesSee cybersecurity risk as the#1 roadblock to generative AI adoptionIdentified security as the#1 priority for generative AI use casesIBM TechXcha

3、nge|2025 IBM Corporation3Maturity of adoptionAs the value of AI continues to grow,so will the risksIBM TechXchange|2025 IBM Corporation4Attackers will target AIAI should be treated as a new attack surface,with new detection and response strategies required for model evasion,extraction,inference and

4、poisoningPrompt injection can drop defenses preventing generation of unwanted material,plus access to exploitable integrations and a wealth of sensitive training dataMalicious models can be uploaded to open repositories,with hidden behavior triggered long after theyve been deployedIBM TechXchange|20

5、25 IBM Corporation5Some emerging threats look familiar while others are newNew world trendsIBM TechXchange|2025 IBM Corporation6Risks are growing every day7Training data risksTransparencyTransparency Lack of training data transparency-Amplified Uncertain data provenance-AmplifiedData lawsData laws D

6、ata usage restrictions-Traditional Data acquisition restrictions-Amplified Data transfer restrictions-TraditionalPrivacy Personal information in data-Traditional Data privacy rights alignment-Amplified Reidentification-TraditionalFairness Data bias-AmplifiedIntellectual property Data usage rights re