Safety
Amazon Introduces Policy and Lambda Interceptors for Secure AI Agents in Bedrock AgentCore
Amazon Bedrock AgentCore now supports policy-based access control and Lambda interceptors to secure AI agents, enabling dynamic validation and governance for enterprise workflows.
Image: AWS Machine Learning
Amazon has introduced new security mechanisms for AI agents through its Amazon Bedrock AgentCore gateway, allowing enterprises to manage secure access to tools across their organization. As companies increasingly adopt AI agents to automate workflows, they face challenges in scaling secure access to thousands of Model Context Protocol (MCP) tools. These tools span different teams, organizations, and business units, creating governance issues that traditional applications do not address. The dynamic nature of AI agent workflows, which involve runtime decisions about tool invocation, complicates auditing and access control. To address this, Amazon Bedrock AgentCore now supports two complementary mechanisms: Policy for deterministic access control and Lambda interceptors for dynamic validation. These tools help build a layered security architecture for agentic solutions. In a demonstration, Amazon used a lakehouse data agent to show how these mechanisms can be combined for geography-based access control. The solution involves a Streamlit UI that authenticates users through Amazon Cognito and passes JSON Web Tokens (JWT) to the agent. The MCP Server exposes five tools, and role-to-tool access is managed through Amazon DynamoDB. AWS Lake Formation enforces row-level and column-level security at query time. *Source: [awsml](https://aws.amazon.com/blogs/machine-learning/secure-ai-agents-with-policy-and-lambda-interceptors-in-amazon-bedrock-agentcore-gateway/)*
Key points
- Amazon Bedrock AgentCore now supports policy-based access control and Lambda interceptors to secure AI agents.
- Policy in Amazon Bedrock AgentCore uses Cedar, a declarative policy language, to enforce deterministic access control.
- Lambda interceptors allow custom code to run before or after each tool call, supporting dynamic validation and payload enrichment.
- The lakehouse data agent demonstrates how Policy and Lambda interceptors can be combined for geography-based access control.
- AWS Lake Formation enforces row-level and column-level security at query time based on user roles and data access permissions.
- The Policy Engine evaluates each tool call against defined policies before permitting access, with deny-by-default semantics.
- Cedar policies can restrict which tools a role can invoke and block access to sensitive operations for certain user groups.