Agentic AI Standards Inventory & Gap Analysis

WORKING DRAFT · June 2026   W3C AI Knowledge Representation Community Group (AI KR CG)  ·  structured by the Knowledge-Domains taxonomy

WORKING DRAFT — this document is not a finished reference but a working memo containing information that needs to be trimmed, evaluated, refined, and maintained: it tracks a dynamic, emergent space and will go stale unless kept current. Treat every entry as a starting point for discussion, contribute via the public mailing list.
Invitation to contribute. You are invited to review this inventory, correct the coverage assessments, add standards missed, and discuss on the public mailing list of the AI KR CG: public-aikr@w3.org — archives at lists.w3.org/Archives/Public/public-aikr/. Anyone may read or write to this list.
Provisional — to be confirmed and maintained. The agentic-AI environment changes rapidly. Every entry, coverage call, and link here is provisional and must be verified and kept current. Coverage tracks ratified standards: GAP none · PARTIAL de-facto / draft only · COVERED ratified standard · ACTIVE bodies engaged. Items whose source is unverified are marked to confirm.

Coverage at a glance

Across the eight knowledge domains, the majority remain a gap or only partially served; standards activity is concentrated in communication/identity, commerce, security, and regulation.

Knowledge domainCoverageSummary
1. Agent types & autonomyGAPNo ratified standard describes agent types or autonomy levels; only governance frameworks reference risk tiers.
2. Cognition & capabilityGAPPlanning, goal specification and perception have no standards; agent memory now has an emerging W3C community group.
3. Interaction, communication & identityPARTIALA vibrant de-facto protocol layer and maturing identity primitives, but no single ratified web standard yet.
4. Commerce & settlementPARTIALSeveral competing de-facto payment protocols; no ratified settlement standard.
5. Oversight, safety & interpretabilityPARTIALStrong de-facto security frameworks and regulatory anchors exist; no ratified oversight vocabulary, and interpretability remains research.
6. Regulation & governancePARTIALBinding regulation exists (EU AI Act, DORA, MiCA); agentic-specific governance is still forming.
7. Agentic standards & standards bodiesACTIVEMany standards bodies are now actively engaged — the landscape is filling, mostly with community groups and de-facto protocols.
8. Applied domainsGAPNo agentic-AI-specific standards; applied work relies on each field's existing domain standards.

1. Agent types & autonomy GAP

No ratified standard describes agent types or autonomy levels; only governance frameworks reference risk tiers.

Standard / effortWhat it coversCoverageOpennessSource
(none dedicated)No standard for agent type / autonomy-level descriptionGAP
EU AI Act risk tiersRisk classification referenced for autonomyRegulationopenReg (EU) 2024/1689
NIST CAISI AI Agent Standards InitiativeUS program; may produce autonomy/agent descriptorsInitiativeopenNIST

2. Cognition & capability GAP

Planning, goal specification and perception have no standards; agent memory now has an emerging W3C community group.

Standard / effortWhat it coversCoverageOpennessSource
W3C AI Agent Memory Interoperability CGPortable agent memory specificationProposed (18 May 2026)openW3C
Planning / goal specificationNo dedicated standards venueGAP
PerceptionNo dedicated standards venueGAP

3. Interaction, communication & identity PARTIAL

A vibrant de-facto protocol layer and maturing identity primitives, but no single ratified web standard yet.

Standard / effortWhat it coversCoverageOpennessSource
A2A (Agent2Agent)Inter-agent communication / coordinationDe-factoopenLinux Foundation
MCP (Model Context Protocol)Agent-to-tool connectivityDe-factoopenAnthropic / LF
ACP (Agent Communication Protocol)REST-native inter-agent messagingDe-factoopenAGNTCY / LF
ANP (Agent Network Protocol)DID-based decentralised discoveryWhite PaperopenW3C
OASF (Open Agent Schema Framework)Agent capability metadata schemaDraftopenAGNTCY
NANDAGlobal agent discovery / registryResearchopenMIT Media Lab (Project NANDA)
DID (Decentralized Identifiers)Agent identityCandidate Rec (Mar 2026)openW3C
Verifiable Credentials 2.0Agent credentialsRecommendationopenW3C
W3C AI Agent Protocol CGInter-agent comms / identity / metadataPre-specopenW3C

4. Commerce & settlement PARTIAL

Several competing de-facto payment protocols; no ratified settlement standard.

Standard / effortWhat it coversCoverageOpennessSource
x402HTTP-402 agent/stablecoin paymentsDe-factoopenx402 Foundation / Coinbase
AP2 (Agent Payments Protocol)Mandate-based agent payment authorisationDe-factoopenGoogle / Coinbase
ACP (Agentic Commerce Protocol)Agent-initiated checkoutDe-factopartly gatedStripe / OpenAI / Meta
ERC-8004 (Trustless Agents)On-chain agent identity / reputationDraft EIPopenEthereum

5. Oversight, safety & interpretability PARTIAL

Strong de-facto security frameworks and regulatory anchors exist; no ratified oversight vocabulary, and interpretability remains research.

Standard / effortWhat it coversCoverageOpennessSource
OWASP Agentic Security Initiative + Top 10Agentic threat taxonomy and controlsDe-factoopenOWASP
CSA MAESTROSeven-layer agentic threat modelDe-factoopenCloud Security Alliance
Five Eyes joint guidanceSecure agentic-AI deploymentGuidanceopenCISA et al.
EU AI Act Art. 14 / prEN 18283Human oversightRegulation / draft stdopenEU / CEN-CENELEC
InterpretabilityNo standard; active researchGAP

6. Regulation & governance PARTIAL

Binding regulation exists (EU AI Act, DORA, MiCA); agentic-specific governance is still forming.

Standard / effortWhat it coversCoverageOpennessSource
EU AI Act (incl. GPAI)Horizontal AI regulationIn forceopenReg (EU) 2024/1689
DORAFinancial-sector operational resilienceIn forceopenReg (EU) 2022/2554
MiCACrypto-asset marketsIn forceopenReg (EU) 2023/1114
GENIUS ActUS stablecoin statuteStatuteopenPublic Law 119-27
NIST CAISI AI Agent Standards InitiativeUS agent standards programInitiative (Feb 2026)openNIST
IMDA Model AI Governance Framework for Agentic AISingapore agentic governanceFrameworkopenSingapore IMDA

7. Agentic standards & standards bodies ACTIVE

Many standards bodies are now actively engaged — the landscape is filling, mostly with community groups and de-facto protocols.

Standard / effortWhat it coversCoverageOpennessSource
W3C AI-KR CGKnowledge representation for AI (this group)CGopenW3C
W3C AI Agent Protocol CGAgent communication / identityCG (pre-spec)openW3C
W3C AI Agent Memory Interoperability CGAgent memoryCG (proposed)openW3C
Linux Foundation Agentic AI FoundationStewards MCP and A2AFoundationopenLinux Foundation
AGNTCY collectiveInternet of Agents (ACP, OASF)CollectiveopenAGNTCY
C2PAContent provenance (agent-generated content)Standard (v2.3)openC2PA
CEN-CENELEC JTC 21EU harmonised AI standards (prEN 182xx)Draft standardsopenCEN-CENELEC

8. Applied domains GAP

No agentic-AI-specific standards exist for this domain; agents operating in a given field reuse that field's own domain standards. There is nothing agent-specific to inventory here — this category is retained only so the eight-domain spine stays complete.

Recommended work items — priority gaps

Where the inventory shows a gap, these are the standards worth developing — carried over from the June 2026 gap analysis and tagged by knowledge domain. The status column notes the state of any effort as of June 2026.

Work itemDomainAddressesStatus of effort (Jun 2026)
Agent Autonomy Level Vocabulary1 · Agent typesReactive / deliberative / learning / goal-seeking agentsOpen; governance anchors only (EU AI Act, NIST CAISI)
Goal Specification Schema2 · CognitionMachine-readable goals & constraintsOpen; no venue
Planning Semantics Vocabulary2 · CognitionPlans, strategies, constraintsOpen; no venue
Agent Memory Model2 · CognitionPortable short / long-term / episodic memoryActive: W3C Agent Memory Interop CG
Agent Communication Protocol3 · InteractionWeb-native agent-to-agent messagingActive (de-facto): A2A, MCP, ACP; W3C Agent Protocol CG
Agent Identity Specification3 · InteractionIdentity / credentials for non-human actorsMaturing: DID v1.1 CR, VC 2.0 Recs
Delegation & Authority Model3 · InteractionScoped, authorised delegationResearch: authenticated delegation
Human Oversight Vocabulary5 · OversightHuman-in-the-loop requirementsOpen; EU AI Act Art. 14 anchor
Creative-Agent Provenance5 · OversightAttribution of agent-generated contentLargely met: C2PA
Policy & Governance Annotations6 · RegulationOperational / legal constraints on agentsFrameworks proliferating; ODRL alignment
Perception Interface Descriptors2 · CognitionSensor / input-modality descriptionOpen; no venue

Companion documents: the taxonomy (category spine) and the Vocabulary. Status: DRAFT — provisional, for community evaluation and maintenance. License: CC BY 4.0.