Safeguarding sensitive data used in generative AI with RAG
summary
Shows a working RAG pipeline built on Amazon S3 → Bedrock embeddings → OpenSearch and identifies five concrete security risk points. Recommends AWS Direct Connect or Site-to-Site VPN, VPC endpoints/PrivateLink, and AWS WAF for network protection. Demonstrates Amazon Macie finding a passport number, bank account and a Korean social security number via regex, and Guardrails for Amazon Bedrock that masks phone numbers and blocks passport numbers; uses IAM and KMS for access control and encryption.
AI-written summary · about
chapters (8)
- 0:00GenAI and RAG overview
- 2:38Simplified RAG architecture
- 3:35Network and traffic security
- 5:58Data protection with Macie
- 7:39Guardrails for Bedrock
- 9:37Data detection demo
- 12:54Guardrails demo
- 15:42Summary and conclusion
related talks
40mSecure Your GenAI Applications
28mImproving responses in generative AI applications
26mBreaking RAG Systems: Exploiting Vulnerabilities & Hardening Your GenAI Applicatio...
41mBuild AI applications with Guardrails for Amazon Bedrock
26mGenerative AI driven sensitive data protection
23mNo RAGrets
55mSecurely build generative AI apps & control data with Amazon Bedrock
40mSecuring Generative AI: Is it all an Illusion?