Amazon Bedrock AgentCore Gateway has been extended with new capabilities to support enterprise deployments of Model Context Protocol (MCP) servers, including enhanced tool schema support, dynamic listing, streaming, session management, and OAuth 2.0 delegated authentication, providing centralized governance, security, and observability.

This post demonstrates how to secure AI agents using Policy (Cedar) for deterministic access control and Lambda interceptors for dynamic validation in Amazon Bedrock AgentCore Gateway. Using a lakehouse data agent example, it shows how to combine both mechanisms for layered security, including geography-based access control.

This post addresses key risks in designing an agentic payment system, such as runaway spending, lack of user consent, credential compromise, and exposure of payment instruments, and shows how Amazon Bedrock AgentCore Payments mitigates them with infrastructure-layer guardrails.