OpenAI Confirms Standards-Body Talks With Anthropic and Google DeepMind

Coffee Summary

  • FACT: On Sep 15, 2026, OpenAI confirmed to CNBC it has been engaging Anthropic and Google on how to collaborate on AI safety; talks have continued since Demis Hassabis’s July proposal for a U.S.-led “Standards Body.”
  • FACT: Hassabis’s proposal framed a public-private or self-regulatory body with federal oversight, often compared to FINRA, to test advanced systems before broad release.
  • FACT: OpenAI global affairs lead Chris Lehane said engagement with Anthropic and Google had been underway for several weeks and that OpenAI does not see a need for an antitrust waiver to coordinate on safety (Bloomberg reporting).
  • CLAIM: Meta CEO Mark Zuckerberg opposed a national FINRA-style AI regulator in an August call with President Trump, favoring lighter voluntary oversight (Business Insider, citing a senior White House official) — Meta is not part of the OpenAI–Anthropic–Google talks as confirmed here.
  • OPINION: Builders should treat this as exploratory industry coordination, not a launched standards body with binding tests or timelines.

What happened

OpenAI confirmed on Tuesday, September 15, 2026, that it has been talking with Anthropic and Google about joint work on AI safety — including the idea of an industry standards body. A spokesperson told CNBC discussions have been ongoing since Google DeepMind chair Demis Hassabis released a July proposal for a U.S.-led “Standards Body.”

Hassabis’s essay argued for a public-private partnership or self-regulatory organization with federal oversight, analogous to the Financial Industry Regulatory Authority (FINRA), with substantial industry funding and independent technical expertise to test powerful models before public release.

France 24 and Bloomberg reported the same core confirmation: OpenAI says it is working with Anthropic and Google DeepMind on safety coordination rooted in that proposal. Bloomberg quoted Chris Lehane saying the engagement had run for several weeks and that OpenAI does not believe an antitrust waiver is required for safety coordination.

Separately, pressure on frontier labs has spiked: Anthropic CEO Dario Amodei recently urged slowing the pace of the most advanced models; Hassabis and OpenAI CEO Sam Altman publicly endorsed the direction of that call (per CNBC). That context explains why rivals are willing to talk — not that any body has been chartered.

Why it matters

If a FINRA-style body ever lands, builders face a new compliance surface: pre-deployment testing, shared evaluation norms, and possibly industry-funded compute for large-scale evals. Even without a charter, voluntary “industry-led standards” (language OpenAI’s Lehane used earlier this month) can still shape procurement questionnaires, insurer underwriting, and enterprise RFPs.

Rivalry does not disappear. OpenAI, Anthropic, and Google remain fierce competitors. Coordination on safety is narrow and contested. Meta’s reported opposition to a national FINRA-style regulator (CLAIM, secondary reporting) shows the industry is not aligned — open-weight release cadence and testing gates can conflict.

What changed

  • Confirmation, not launch — OpenAI on-record confirmation of talks; no announced charter, board, test suite, or go-live date.
  • Hassabis proposal as catalyst — July essay remains the stated origin of the discussion frame (FINRA analogy, U.S.-led, industry-funded testing).
  • Antitrust posture — Lehane: no waiver needed for safety coordination (FACT via Bloomberg); legal risk for competitors still exists if talks wander into pricing or market allocation.
  • Policy stack — Lehane previously said industry-led standards would complement, not replace, mandatory federal safeguards; OpenAI has also signaled support for independent verification concepts in pending legislation (CLAIM/policy narrative — verify bill text separately).
  • No consensus — Multiple outlets note there is still no agreement on structure or next steps.

Who should care

  • Policy, trust & safety, and compliance leads at companies shipping or hosting frontier models.
  • Enterprise buyers writing AI vendor questionnaires (ask what voluntary standards each lab will accept).
  • Open-weight and open-source maintainers watching whether pre-release testing norms favor closed labs.
  • Antitrust counsel monitoring competitor-to-competitor safety forums.
  • Investors pricing regulatory friction into AI infrastructure timelines.

Limitations

  • Talks are ongoing; this is not a launched Standards Body.
  • Google and Anthropic did not immediately comment to CNBC; OpenAI’s confirmation is the primary on-record pillar for Sep 15.
  • Meta opposition details come from secondary White House-sourced reporting — treat as CLAIM, not OpenAI’s statement.
  • No public draft charter, membership list, fee schedule, or evaluation protocol.
  • Amodei’s slowdown call and Altman/Hassabis reactions are adjacent politics, not proof the standards body exists.

What to do next

  • Bookmark OpenAI / Anthropic / DeepMind policy blogs for any charter draft or voluntary standard text.
  • Add a vendor question: “Will you submit frontier models to third-party or industry pre-deployment testing if a FINRA-style body forms?”
  • Separate safety coordination from product roadmap in board briefings — do not assume slower release cadence until written commitments appear.
  • Track Meta and other labs’ positions; industry split affects whether standards become de facto requirements.
  • Have counsel review any multi-lab working group invitations for antitrust hygiene (agenda limits, minutes, no commercial topics).
  • Do not redesign compliance programs around a body that has not been created.

AIImpish Take

September 15’s news is confirmation of talks, not the birth of AI’s FINRA. Hassabis supplied the template; OpenAI supplied the on-record nod; Anthropic and Google remain in the frame without a public charter. Builders should watch for concrete artifacts — test scopes, funding, federal oversight language — and ignore breathless “AI regulator launched” headlines until those exist.