AWS re:Inforce 2025 - Building secure RAG applications on Amazon Bedrock (APS322)

Riggs Goodman and Ismail Manuel from AWS discuss using Retrieval Augmented Generation (RAG) with vector databases to add context to generative AI applications like Amazon Bedrock. Riggs provides an overview of Bedrock, emphasizing its security and compliance features, and explains how RAG architectures differ from traditional database applications in terms of security. Ismail then details three strategies for building secure RAG architectures: authorization at the application layer, using separate knowledge bases for different user groups, and employing metadata filtering for fine-grained access control, ensuring that sensitive data is only accessible to authorized users and protecting against prompt injections.