Fine-grained access control for Amazon Bedrock Managed Knowledge Base tools

This article demonstrates how to implement fine-grained access control for shared AI agents using Amazon Bedrock Managed Knowledge Base, ensuring each user can only search documents they're authorized to view.

• Use ACL awareness in Managed Knowledge Base to return only documents the authenticated user is allowed to view
• Inject verified user identity from JWT tokens into tool calls via request interceptors on AgentCore Gateway
• Validate injected userContext against JWT claims using Cedar policies in Policy in AgentCore
• Restrict knowledge base access with resource-based policies to deny searches from unauthorized principals
• Combine multiple defense layers to prevent LLM spoofing and unauthorized access attempts
• Deploy sample architecture with AWS CDK supporting multi-tenant scenarios and department-level access control

The solution prevents unauthorized document disclosure in multi-tenant environments by ensuring the LLM cannot decide user identity, with verified credentials injected at the gateway layer and validated at both tool and resource layers.