Secure AI agents with Policy and Lambda interceptors in Amazon Bedrock AgentCore gateway

This article demonstrates three design patterns for securing AI agents in Amazon Bedrock AgentCore using Policy and Lambda interceptors.

• Policy provides deterministic, auditable access control using Cedar policy language
• Lambda interceptors enable dynamic validation, token exchange, and payload transformation
• Design 1: Policy-only restricts tools by role with forbid/permit rules
• Design 2: Interceptor-only exchanges JWT for tenant-scoped IAM credentials (act-on-behalf)
• Design 3: Combined approach uses interceptor to inject geography, policy to enforce data-residency rules
• REQUEST interceptors run before Cedar policies, enabling request enrichment before evaluation
• RESPONSE interceptors filter tool lists and redact sensitive information dynamically
• Use Policy for deterministic rules; use interceptors for external lookups and transformations
• Sample lakehouse agent demonstrates role-based access (policyholders, adjusters, admins)
• AWS Lake Formation enforces row/column security at query time
• Full implementation available in GitHub repository with CDK deployment scripts

The layered security architecture combines Policy's auditability with interceptors' flexibility to build comprehensive agent governance at scale.