A walkthrough of security and trust challenges in agentic AI systems, covering credential replay attacks, rogue agents, user impersonation, and overpermissioning. Key mitigations discussed include using TLS/mTLS for secure communication, authenticating agents via an identity provider, delegation tokens that bind user and agent identities together, token exchange at each hop to validate propagation, scope restriction via least privilege, and a secure vault for last-mile credential management between MCP servers and tools.
•12m watch time
Sort: