Kam Sween

This talk addresses the critical security challenges arising from increasingly capable AI agents that perform actions in the real world. It emphasizes the need for robust identity and access control mechanisms to prevent issues like secrets in prompts, overly broad scopes, and difficult troubleshooting. The core thesis is that by adopting open standards and implementing proper identity management, developers can build more secure and trustworthy AI agent systems.