Guardian agents monitor AI agent identities at runtime to reduce risks from inherited permissions, stale credentials, and prompt injection.