Harnessing Generative AI: Security strategies for next-gen applications
summary
Claims a generative security scoping Matrix that classifies five scopes (buy consumer, buy enterprise, build on pre‑trained models, fine‑tune, train from scratch) and asserts no radical rethink of security is needed while calling out new risks. Implements a working scope‑3 demo using Amazon Cognito, S3, OpenSearch, CloudWatch, WAF and Amazon Bedrock guardrails to block out‑of‑scope answers (investment advice) and log model invocations. Notes attackers use prompt injection, model DoS, data poisoning and model theft and that almost nobody is fine‑tuning or training foundation models yet.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 1:26Security for GenAI adoption
- 3:30GenAI capabilities basics
- 5:17New security approach for GenAI
- 6:24Generative Security Scoping Matrix
- 8:12Security domains Scope 1 & 2
- 10:17Governance for consumer apps
- 13:26Controls & resilience Scope 1 & 2
- 14:29Security for Scope 3 apps
- 16:17Risk management Scope 3
- 18:47Demo setup & architecture
- 20:55Live demo Scope 3 app
- 22:20Security in Amazon Bedrock
- 23:22Logging & monitoring GenAI workloads
- 23:44Amazon Bedrock guardrails
- 25:29Security Scope 4 & 5
- 26:53Key takeaways
related talks
38mAccelerate securely: The Generative AI Security Scoping Matrix
44mGenerative AI Security — A Practical Guide to Securing Your AI Application
19mSecurity controls for generative AI use cases
40mSecuring Generative AI: Is it all an Illusion?
20mSecuring generative AI: Privacy and compliance considerations
19m5 ways generative AI can enhance cybersecurity
27mSafeguarding Generative AI Applications: Navigating Security Challenges and Best Pract...
09mSecuring Generative AI Workloads