Agent ID

Editorial note

The standards conversation that everybody knows is coming and nobody has yet been able to convene with sufficient authority. This is the placeholder draft.

Our analysis

Agent ID is the standards conversation everybody in the category knows is coming and that no working group has yet been able to convene with sufficient cross-vendor authority. The current draft (0.2, published as a joint effort of contributors from Anthropic, OpenAI, and several large infrastructure vendors but not formally hosted) sketches a primitive set: an agent identifier (akin to a service account), a principal-on-behalf-of claim, a capability assertion (what the agent is currently authorized to do, scoped and time-limited), and an audit envelope. The model is roughly the OAuth analog, with the parts that do not transfer cleanly clearly marked. The hard problem is that identity standards are political in a way that tool protocols are not — they touch on liability, regulatory compliance, and the question of who underwrites an agent's actions when it acts on behalf of a human. We expect 2027 to be the year this conversation either lands a draft into a serious working group or collapses into vendor-specific implementations that get harmonized later. The Review will track this closely.

What we are watching

• Whether the draft gets hosted under a real working group (IETF? W3C? OpenID Foundation?)
• Vendor-specific identity primitives shipping in 2026 H2 that pre-empt the standard
• Regulatory pressure (EU AI Act, US executive orders) that could force convergence

How to cite this entry

Citation format: The Agentic Review. "Agent ID — Agent Identity Working Draft." Standards Tracker, last updated 2026-05-18. https://agentic.review/standards/agent-id/. See our citation style for additional guidance.

Update history

Material updates to this entry are dated and logged in our corrections log. The version above reflects the most recent review by the standards desk.