Wall Street Already Has the AI Regulator

Google DeepMind CEO Demis Hassabis just asked Washington to copy Wall Street's oldest trick — let the industry fund its own regulator, then make compliance mandatory once the playbook works.

What's new: On July 14, Hassabis published "A Framework for Frontier AI and the Dawning of a New Age," a personal manifesto proposing a U.S.-led standards body modeled on FINRA — the industry-funded watchdog that polices brokerages under SEC oversight.

In an exclusive Axios interview, he said the body should be operational "before year-end."

The quote:

"We've essentially found a way to make sand think," Hassabis writes. "It's miraculous."

He told Axios today's AI cyber risks are "warning shots" — and that within 18 months, graver biological and nuclear capabilities could live in open-source models "beyond any government's control."

"What we collectively do now," he writes, "will determine how the next phase of civilization unfolds."

How it would work:

• Voluntary first: Frontier labs share models up to 30 days before release for safety testing on cyber, biological, and deception risks.
• Mandatory later: Once audits prove "effective and robust," clearing the body becomes a requirement for U.S. deployment.
• Industry-funded: Labs pay the bills — Hassabis argues that's the only way to hire top evaluators and buy the compute for large-scale red teams.
• Independent board: Majority-independent directors — Turing Award winners, credentialed experts — plus industry, government, and open-source seats.
• Quarterly benchmarks: Thresholds update as capabilities evolve; the body eventually builds held-out tests labs can't train toward.
• Circuit breaker: The body could coordinate an industry-wide slowdown if dangers mount.
• Global scope: Rules apply to frontier models "no matter their country of origin or whether they are open or closed."

The prestige play:

Hassabis told Axios the "frontier" designation would become a badge: "I think that's a pretty nice, prestige kind of asset to have."

Being tested means you matter.

The wake-up call:

Washington's improvised crackdown on Anthropic's Mythos and Fable models was "a bit of a wake-up call," Hassabis said — proof ad hoc export-control orders are no substitute for rules.

• Anthropic saw its most powerful models frozen overnight, then spent 2½ weeks negotiating release with no established protocol.
• OpenAI restricted GPT-5.6 Sol to government-vetted partners at launch, then released publicly last week only after Commerce Department testing and negotiations.

Both labs got through. Neither had a playbook.

The FAA vs. FINRA split:

Anthropic CEO Dario Amodei wants binding government regulation modeled on the Federal Aviation Administration — mandatory third-party testing, government power to block unsafe deployments.

Amodei writes: "Frontier AI models, like airplanes, should be required to go through technical testing and auditing, and their release should be blocked or reversed as a threat to public safety if they do not meet high standards of safety."

Hassabis wants industry to fund and staff the audits, with Washington holding veto power from the sidelines.

The lab chiefs behind Gemini and Claude now agree Washington should regulate them. They disagree on who writes the exam.

Why FINRA:

Lawfare argued in March that finance already solved the problem AI faces — information asymmetry, rapid innovation, and a race to the bottom on safety.

FINRA governs 3,000+ broker-dealers on a $1.3 billion industry-funded budget. Congress gave it antitrust-safe coordination powers competitors can't replicate voluntarily.

Every major frontier lab except xAI already belongs to the Frontier Model Forum — a proto-SRO without statutory teeth.

Hassabis spent months briefing the Trump administration, fellow lab leaders, and European officials before going public.

"The noises I've been hearing are very positive," he told Axios. Other lab leaders agree at a high level: "This is where the industry needs to go."

The timing:

Hassabis believes AGI is "probably only a few short years away" — and that we're standing in "the foothills of the singularity."

The EU AI Act's August 2 enforcement date still stands for GPAI fines and Article 50 transparency — even after Omnibus delayed high-risk Annex III rules to December 2027. Japan enacted election deepfake rules July 13. Canada's OSFI named Mythos in bank inboxes the same week.

Every jurisdiction is improvising. Hassabis is offering a template.

Convina's view: Hassabis isn't asking for less regulation — he's asking for regulation labs can afford to lose at. A FINRA-style body lets incumbents write the first draft of the benchmarks, fund the evaluators, and sell "frontier-class" certification as a marketing badge while Washington keeps the kill switch it already proved it has. That's smarter politics than Amodei's FAA — and exactly why you should read the fine print on who sets the thresholds. The Mythos scare showed Commerce can pull models offline overnight; Hassabis's proposal would make that power routine, industry-sanctioned, and harder to challenge in court. If you're betting on IPO-week labs, price the compliance desk before you price the compute.