The Rivals Want a Referee They Can Afford

The three labs racing hardest to build superhuman AI just published the same prescription in three different fonts — independent testing, a U.S.-led standards body, and restricted access to the most dangerous models — while Washington is already flipping the kill switch without asking which blueprint wins.

What's new: Axios reported on July 16 that Demis Hassabis, Sam Altman, and Dario Amodei each dropped detailed governance memos over five weeks — the same stretch in which the Trump administration twice restricted frontier releases.

• Hassabis (July 14): A FINRA-style, industry-funded standards body that starts with voluntary 30-day pre-release reviews and could harden into a mandatory U.S. market gate.
• Amodei (June 10): An FAA-style federal regime with binding third-party tests and government power to block deployment on Day 1.
• Altman (July 1, Financial Times): A U.S.-led IAEA-style international forum that certifies countries and companies, using market access as leverage.

All three break from the old self-reporting era. None wants a broad AI crackdown. All cite cyber and bioweapon risk. All want America setting the terms.

The rare truce: Hassabis's July 14 manifesto drew endorsements rivals rarely hand each other.

• Sam Altman called it "thoughtful."
• Satya Nadella called it "an important piece" toward "a frontier ecosystem that promotes innovation and choice."
• Elon Musk — who joked in June about naming a regulator "AIAIAI" — wrote it was "a good starting point for discussions."
• Anthropic co-founder Jack Clark went further: "At this point, everyone at the frontier of AI agrees that third parties should test out AI systems and use these to develop standards to feed into policy."

Axios reports Meta's Mark Zuckerberg is drafting his own memo.

Washington didn't wait: The administration publicly rejects an "FDA for AI" — White House AI advisor Sriram Krishnan said in July there "will not be an FDA for AI." Privately, officials already improvised twice.

• June: Anthropic's Claude Mythos 5 and Fable 5 went dark for weeks after national-security concerns.
• July: OpenAI limited GPT-5.6 Sol to "trusted partners" at the government's request, per CNBC.
• July 17: The White House launched Gold Eagle — a clearinghouse CNBC sources say will greenlight which companies access frontier models.

Hassabis told Axios he wants his standards body running in "months," ideally before year-end. Gold Eagle is already live.

Where they split: The fight is over who holds the veto.

• Amodei wants a federal agency that can block a release immediately — modeled on aircraft certification.
• Hassabis wants industry-funded self-regulation under federal oversight — voluntary first, mandatory once the protocol "is shown to be effective and robust."
• Altman wants an international forum that trades compliance for access — the nuclear-agency playbook.

Three regulators. One frontier. No agreement on the key.

The catch: Axios flags what skeptics see immediately — OpenAI, Google, and Anthropic already have the lawyers, security teams, and Washington relationships to survive a heavy certification process. Startups and open-source developers do not.

Critics call it regulatory capture: safety rules that entrench the incumbents who wrote them. The timing makes the suspicion harder to dismiss — Moonshot's Kimi K3, a 2.8-trillion-parameter open-weight model rivaling Claude Fable 5 and GPT-5.6 Sol, shipped the same week Washington tightened the proprietary guest list. A FINRA for closed models does not gate a GitHub download.

Convina's view: The Wild West is over because the people with the most compute decided it should be — and they are offering Washington three polished blueprints for the lock they already built. Enterprise buyers should assume the guest list hardens before the standards body opens. Procure for sovereignty, not for whichever referee the labs agree on next.