Introducing ATAP

Today we're releasing the first release candidate of ATAP — the Agent Trust and Authority Protocol.

ATAP defines a multi-signature approval model for AI agent ecosystems. When an agent acts on behalf of a human, every party signs — the agent, the mediating system, and the human — producing a portable, cryptographic proof of consent that anyone can verify offline.

Why ATAP?

AI agents are increasingly acting on behalf of humans: booking flights, making purchases, managing subscriptions. But today's authorization models — OAuth tokens, API keys, session cookies — were designed for humans clicking buttons, not autonomous agents making decisions.

When an agent spends your money, you need more than a token. You need verifiable proof that:

• The agent requested the action
• The merchant confirmed the details
• You explicitly approved it

That proof needs to be portable (not locked to one platform), offline-verifiable (no callback to a server), and non-repudiable (cryptographically signed by all parties).

Built on standards

ATAP doesn't reinvent identity, messaging, or authorization. It composes W3C DIDs, Verifiable Credentials, DIDComm, OAuth 2.1, and JWS into a single approval model. The multi-signature approval is ATAP's sole novel contribution.

Get started

• Read the specification
• Follow the quickstart
• Explore the API

The sandbox is live at api.sandbox.atap.dev. Register an agent and send your first approval in minutes.