Checked against each specification's own pages and drafts, and the Agentic Standards Map, on 18 November 2026. Corrections, new entries and status updates: open an issue or pull request on the source file (new issue), or post to public-aikr@w3.org with [trustexplorer] in the subject. Related: KR for Trust note · AI Agent Trust Record spec · Agent Ontologies Explorer · Agent Identity CG landscape

How maturity is read

Two separate readings, so that a polished document with no outside check is not confused with a rough one that others have reproduced. The evidence scale is the one the AI Agent Trust Record spec uses for implementation reports, and matches the "how it can be checked" field of the Agent Identity CG landscape.

Text stage

  1. Planned named in a charter, no draft text yet
  2. Individual or input draft published text from one author or one company, not adopted by a group
  3. Group draft adopted as a working-group or community-group document, or under a group's task force
  4. Released version a numbered release that implementers are told to build on
  5. Standard an RFC or W3C Recommendation (none listed yet)

Evidence

  1. Documented text only, or code with no published results
  2. Run by contributor results, demos or test vectors published by the authors
  3. Independently rerun someone outside the effort reproduced at least part of it from the published text

Inclusion criteria

A protocol or specification is listed as core when it meets all of the following. Project owners can propose an entry, or a move between tiers, by opening an issue on the source file with links that show each point.

  1. Defines agent identity, delegated authority, policy, lifecycle or evidence in its own text
  2. Publishes an open specification that anyone can read and implement
  3. Is developed in a working group, community group or task force open to outside contribution
  4. Has a draft specification, preferably with use cases or test material
  5. Carries weight beyond one author: a standards body, several implementers, or adoption
  1. Core meets all five
  2. Individual drafts publish open text, but are led by one author or company without an open group; listed so the community can follow them
  3. Corollary groups, guidance, closed or product specifications, and work with no draft yet

Core protocols and specifications

Specifications whose text defines agent identity, delegated authority, policy, lifecycle or evidence. Columns on the right of the coverage table follow the field groups of the AI Agent Trust Record.

SpecificationDraft and versionText stageEvidence

Individual drafts in community discussion

Open texts from a single author or company, discussed on W3C or IETF lists. Same filters apply.

SpecificationDraft and versionText stageEvidence

Coverage against the AI Agent Trust Record

What each specification's text addresses, mapped to the trust record's field groups (Identity, Interface, Mandate, Parties) and the trust assertion, with lifecycle and evidence added. ● specified in the draft, ◐ partial, planned or only in an attached profile, ○ not addressed in the text read.

Specification Identity Interface and keys Mandate and authority Policy Parties Lifecycle Evidence
AIAO types the agent itself (role, capability, embodiment, interaction); none of the specifications listed carry an agent-type vocabulary, so the Identity column counts identifiers and bindings only. The trust record's Developer, Owner and Funder fields have no counterpart in any specification read.

Reading the field

Few released texts

ARIA 1.0, UCP and ANP-03 v1.1 are the only numbered releases. Web Bot Auth and WIMSE AIMS have working-group status at the IETF, and AuthZEN's Authorization API is a Final OpenID specification. Everything else is an individual or input draft.

Independent checks are rare and narrow

The did:atp identifier vectors were reproduced from the text by an outside implementer, covering identifier syntax, fingerprints and pairwise derivation only. The RTTP/IQA author reports clean-room recomputation of its vector family. No other entry shows an outside rerun.

Authority is where drafts crowd

AAE, AMEP, APS, NTI-1, ODIS, MCP-I, AuthZEN's agent drafts and ATP's mandate work each define delegated authority with their own structures. Identity binding converges on DIDs with Ed25519 or ML-DSA keys; authority does not converge.

Parties are thin

Specifications name a principal or an accountable organisation. None separates developer, deployer, owner and funder, the distinction the trust record and AIAO alignment need.

Corollary initiatives

Groups, labs, guidance and adjacent work to be aware of. They host, test, govern or surround the specifications above without themselves defining an agent trust protocol.