The Right Way to Secure AI Agents with Code Examples - IAM441

Learn how to implement secure identity management and authorization patterns for generative AI agents in this conference talk from AWS re:Inforce 2025. Explore the unique security challenges that arise when AI agents operate autonomously across on-premises environments and multiple cloud providers, including identity authentication, propagation, delegation, and resource authorization complexities. Discover how to leverage Amazon Cognito's OAuth2-based identity management system for machine-to-machine authentication, combined with Amazon Verified Permissions for fine-grained authorization controls to create secure delegation patterns for AI agents. Master techniques for preserving human identity and consent, maintaining agent machine identity, and retaining essential request context throughout complex agent chains. Examine real-world implementation examples featuring agents built on Amazon Bedrock and other popular frameworks, with practical code demonstrations showing how to architect secure agentic AI solutions that maintain proper security boundaries while enabling seamless cross-environment operations.