An open standard for authorizing AI agents.

Agents are already in production. They're reading and writing files, calling APIs, sending emails, managing cloud infrastructure — most of them running with whatever API key was in the environment at startup, logging nothing useful, with no revocation mechanism beyond "kill the process."

This is how the web looked before session management was standardized. Every team improvising, every implementation slightly different, every system insecure in its own way. IAM addressed this for human identities and DevOps practice for machine identities. Autonomous agents still lack a shared answer.

The window to establish shared standards is while the ecosystem is still forming. Once every agent framework has built its own identity model and delegation mechanism, interoperability becomes a much harder problem. We aim to do for agent authorization what OAuth did for delegated authorization: a common substrate that everyone builds on, so the hard security work gets done once and done well.