Identity-aware AI data agents with AWS Lake Formation and Trusted Identity Propagation

This article demonstrates how to build identity-aware AI data agents that propagate user identity through every hop from the user to the data layer, enabling Lake Formation to enforce per-user access controls without modifying existing governance policies.

• Configure Bedrock AgentCore Runtime to forward user identity tokens via custom HTTP headers to the agent container
• Use Bedrock AgentCore Gateway metadata configuration to propagate identity headers to Lambda without exposing tokens to the foundation model
• Perform server-side token exchange in Lambda to convert identity tokens into Lake Formation credentials scoped to the real user
• Ensure full CloudTrail audit trails recording the actual user who accessed data through the agent
• Maintain clean separation where the foundation model never sees user tokens while Lake Formation enforces grants against real user identity

This pattern enables secure, governed AI agents querying lakehouses with per-user access controls, full auditability, and no changes to existing Lake Formation policies.