Agent Trust & Identity: Vocabulary Crosswalk Across W3C Community Groups

Abstract

Several W3C Community Groups formed through 2025–2026 independently address AI agent identity, trust, and interoperability, each producing its own terminology. This note is an annotated crosswalk of the vocabulary each group uses, the underlying standards each reuses, and where their scopes visibly overlap or leave gaps. It supersedes the earlier draft of this note, which incorrectly framed the comparison as a mapping onto AIKR's own AIAO classification scheme; AIKR does not currently maintain a vocabulary covering this space, and the note below compares the external efforts on their own terms.

Status of this document

This is a non-normative landscape/crosswalk note draft maintained by AIKR CG's. It records terminology as published in each group's charter, repository README, or mailing-list announcements as of July 2026 — several of these groups are newly formed and their vocabularies are not yet stable.

1. Vocabulary crosswalk

Community Group / Effort	Steward	Distinctive vocabulary	Underlying standards reused	Notes
Agent Identity Registry Protocol CG (`public-agent-identity`)	Proposed by TrustLayer Foundation A.C. (Agent-ARIA spec)	DID method, agent credential format, trust negotiation protocol, trust levels (L0–L3), credential lifecycle / revocation, post-quantum cryptography	W3C DID Core, W3C Verifiable Credentials Data Model; integration profiles for MCP / A2A / OAuth / OIDC / SPIFFE	Only group in this set with a live registry (registry.aria.bar) and published spec (aria.bar/spec) as of July 2026.
AI Agent Protocol CG	W3C CG, broad umbrella (proposed May 2025)	Agentic Web, inter-agent communication protocols, agent identity models, capability metadata, protocol interoperability, orchestration	Verifiable-credential-based trust (unspecified VC profile)	Parent/umbrella scope; identity and security/privacy work items were explicitly left open for other groups to fill — see ATP below.
Agent Trust Protocol (ATP) CG	Formed at invitation of AI Agent Protocol CG chairs	Security/privacy modules: pairwise DIDs, selective disclosure, zero-knowledge proofs	W3C DID Core (pairwise DIDs), ZKP schemes (unconfirmed which)	Direct scope overlap with Agent Identity Registry Protocol CG's trust-level/credential model — both define trust negotiation over DID-anchored identity; no observed coordination between the two as of July 2026.
Agent Declaration and Assurance CG (ADACG)	W3C CG	Agent Declaration, Agent Assurance, Open KYA (Know Your Agent) Manifest, Graduated Conformance Profiles (Core/Deploy/Transact), runtime integrity bindings	Trusted Execution Environments (TEEs), DID	Declares identity/boundaries/accountable-party rather than issuing registry-based credentials — a different mechanism for a similar goal to Agent-ARIA; conformance-profile framing (Core/Deploy/Transact) has no equivalent in Agent-ARIA or ATP.
Semantic Agent Communication CG	W3C CG	Identity binding, capability disclosure, semantic intent modeling, delegation chains, verifiable execution accountability, agent-native interaction models	RDF / Linked Data, DID, Verifiable Credentials	"Delegation chains" here overlaps conceptually with Agent-ARIA's Agent Trust Policy and with ADACG's assurance model; framed semantically (RDF-native) rather than protocol-native, which is itself a gap between the semantic-web and credential-registry camps.
AI Agent Memory Interoperability CG	W3C CG (proposed May 2026)	Portable agent memory, identity-bound memory records	Post-quantum signatures (ML-DSA-65 / FIPS 204) — same primitive family Agent-ARIA uses	Adjacent, not overlapping: reuses Agent-ARIA-adjacent cryptographic primitives for a different object (memory, not identity/authorization).
WebAgents CG	W3C CG	Hypermedia Multi-Agent Systems (hMAS), Web-based MAS, manageable affordances	Linked Data, Semantic Web	No identity or trust vocabulary at all in published interoperability report — a structural gap if hMAS agents are expected to interoperate with any of the identity-bearing agents above.
UHDS CG (health data, adjacent domain)	W3C CG	Modular health data schemas, patient consent/control	Verifiable Credentials, Zero-Knowledge Proofs, Privacy-Enhancing Technologies, federated learning	Not agent-identity-specific, but its ZKP + VC combination is closer to ATP's stated direction than anything ATP itself has published — a candidate reference point.

2. Observed overlaps and gaps

Overlap: Agent Identity Registry Protocol CG and ATP CG both define DID-anchored trust negotiation with no visible coordination between them — the clearest duplication risk in this set.
Overlap: "Delegation chains" (Semantic Agent Communication CG), the Agent Trust Policy (Agent-ARIA), and Graduated Conformance Profiles (ADACG) all attempt to answer the same question — what is this agent authorized to do, and on whose behalf — via three incompatible mechanisms (RDF delegation graph, signed policy document, conformance-tier declaration).
Gap: No group in this set has published a vocabulary for revocation semantics beyond Agent-ARIA's own trust ledger; ADACG's runtime integrity bindings are the closest adjacent concept but are not framed as revocation.
Gap: WebAgents CG's hMAS work has no identity/trust vocabulary at all, despite being a plausible interoperability partner for any of the identity-bearing frameworks above.
Gap: None of these groups has yet published a stable term for cross-registry agent identity resolution — i.e., how a verifier reconciles an Agent-ARIA credential against an ADACG KYA Manifest for the same underlying agent.