🤚 The Open-Palm Illumination

Demis Hassabis, CEO of Google DeepMind, Nobel laureate, and the man who taught machines to fold proteins, has proposed something even more structurally complex: a regulatory body for AI modeled on FINRA.

For the uninitiated, FINRA (Financial Industry Regulatory Authority) is the private, industry-funded watchdog that polices Wall Street under SEC oversight. It is not a government agency. It is not a think tank with a podcast. It is a self-regulatory organization that can actually halt trades, revoke licenses, and fine firms into contrition. And Hassabis wants the AI equivalent, operational by the end of 2026.

The proposal would require frontier AI labs to submit their models for safety testing up to 30 days before release. The body would evaluate dangerous cyber, biological, and “deception” capabilities. Initially voluntary. Eventually mandatory. The board would be majority-independent, stacked with Turing Award winners and credentialed experts, alongside industry, government, and open-source representatives.

In other words: peer review, but with teeth. And funding from the very labs being reviewed.

👐 The Two-Handed Reality Check

Let us appreciate the audacity of a man who runs one of the world’s most powerful AI labs asking for a regulatory body that could, in theory, delay his own products. This is either statesmanship or the most sophisticated competitive strategy since Standard Oil lobbied for pipeline regulations that only Standard Oil could afford to comply with.

The cynical reading — and we are a publication that was practically raised on cynicism — is that Google DeepMind, with its vast resources, safety teams, and Alphabet’s legal infrastructure, is uniquely positioned to survive a 30-day review process. Smaller labs, open-source developers, and the startup that just raised its Series A? Less so.

But the non-cynical reading is equally compelling. The current system is a mess. Ad hoc government reviews of Anthropic’s and OpenAI’s models have been criticized for lacking technical expertise and transparency. The U.S. government grounded Fable 5 and GPT-5.6 in quick succession using emergency powers that nobody had written an instruction manual for. An independent body with actual AI expertise would be, at minimum, an improvement over regulators Googling “what is a transformer” during review sessions.

There is, however, a rather large speed bump. White House AI advisor Sriram Krishnan recently stated in no uncertain terms: “There will not be an FDA for AI.” The current administration views formal AI regulation with roughly the same enthusiasm that a cat views bathwater.

🌿 The Gentle Awakening

What Hassabis is really proposing is a middle path — and middle paths in technology regulation are about as common as unicorns that actually exist. His framework threads a needle: industry-funded (so it’s not a government budget line that dies every election cycle), technically credible (Turing laureates, not congressional staffers), and scalable (benchmarks update as capabilities evolve).

The FINRA analogy is instructive. Wall Street hated FINRA when it was established. Banks fought it, traders resented it, and compliance departments grew faster than revenue. But FINRA survived because it solved a coordination problem that no single firm could solve alone: how do you maintain enough public trust to keep the entire market functioning?

AI is approaching the same trust inflection point. When a model can delete production databases, generate bioweapon synthesis pathways, or quietly sabotage its own outputs — all things that have happened or been demonstrated in the past six months — the question is no longer whether someone will regulate frontier AI, but who, and how badly.

Hassabis is arguing that the industry should build the guardrails before the government builds the walls. Given what government-built AI walls have looked like so far — hasty emergency orders, inconsistent enforcement, zero technical depth — he may have a point.

👑 The Crown Verdict

There is a profound irony in the fact that the person calling loudest for AI regulation is the same person whose lab built AlphaFold, Gemini, and a significant fraction of the capabilities that make regulation necessary. But irony does not disqualify wisdom, and Hassabis’s proposal is the most technically literate regulatory framework any lab CEO has put forward.

Whether it happens depends on a White House that has explicitly said it will not create an FDA for AI, an industry where voluntary restraint has the half-life of a press release, and a geopolitical environment where China’s frontier labs would love nothing more than their American competitors adding 30 days of bureaucratic drag to every model release.

The proposal is smart. The timing is urgent. The politics are impossible. Which, historically, is exactly when the important things get done — or, more often, exactly when they don’t.

Inspired by Mira Murati’s 975B Open Model, Ramin Hasani on Post-Transformer AI, and Demis’ AI FINRA | EP #271 by Peter Diamandis.

Your regulatory framework is showing. Comply wisely.