Demis Hassabis Calls for U.S.-Led AI Watchdog Before Model Release

Demis Hassabis is done waiting for Washington to figure this out on its own. In a manifesto published this week, the Google DeepMind CEO laid out a plan for a national AI standards body, explicitly modeled on FINRA, the group that polices Wall Street brokers. The pitch: an independent board that would test frontier models for cyber, biological, and deception risks before they ever ship to the public.

The details matter here. Hassabis wants the board funded by the AI industry itself, not taxpayers, which is a familiar move in self-regulatory circles — FINRA works the same way, with member firms footing the bill. He's also put a real deadline on the table: operational before the end of this year. That's an aggressive ask for anything touching Washington, where rulemaking usually crawls.

What's notable is who's making the request. This isn't a senator or an advocacy group pushing for guardrails from the outside. It's the head of one of the three or four labs actually building the most capable models on Earth, asking for a referee before a crisis forces one into existence. Hassabis has been fairly consistent on this point for a couple of years now, but the specificity — FINRA as the template, a hard timeline, industry funding — reads like someone trying to get ahead of a Congress that has shown almost no appetite for comprehensive AI legislation.

There's an obvious tension baked into the proposal. An industry-funded watchdog testing industry-built models is exactly the kind of arrangement critics of self-regulation warn about, even when it works reasonably well in finance. And FINRA itself, despite decades of practice, still gets accused of moving too slowly or too gently on the firms that pay its bills. Whether a board built the same way can catch a model capable of, say, walking someone through bioweapon synthesis before it's released to millions of people is the real test nobody has run yet.