# Why OAuth Client ID Metadata Is Essential for Securing Ephemeral AI Agents

Jonas Iggbom of Curity explains how OAuth must adapt to secure AI agents, especially short-lived agents that cannot be pre-registered like traditional applications. He argues that shared clients destroy traceability and revocation control, while dynamic client registration does not scale because it leaves large numbers of temporary clients to manage. The proposed alternative is the client ID metadata document, or SIMD, where an agent presents a URL as its client ID and the authorization server fetches metadata and signing keys at runtime. Trust depends on three elements working together: domain and TLS trust, allow-listed domains, and private-key JWT assertions that bind the token request to the published metadata. For multi-agent delegation, he highlights token exchange with nested ACT actor claims to preserve an auditable chain of who acted on whose behalf, and points to rich authorization requests and per-agent policies to constrain each hop’s audience and scope.