2 hours ago
20
Demis Hassabis, co-founder and CEO of Google DeepMind, wants the US government to build a new regulatory body specifically designed to police the world’s most powerful AI models. In an exclusive interview with Axios, Hassabis laid out a vision for an agency that would be funded by the AI industry itself, staffed by elite technical talent, and ultimately answerable to Washington.
The kicker: this watchdog wouldn’t just write reports. It would have the authority to screen advanced AI systems before release and, if things get dicey, coordinate an industry-wide slowdown.
What Hassabis is actually proposing
The plan calls for a “systematic” approach to AI regulation. Hassabis wants pre-release testing requirements baked into the framework, meaning companies would need to prove their models are safe before shipping them to the public.
This isn’t a sudden pivot for the DeepMind chief. Back in March 2026, Hassabis warned publicly about “race conditions” in AI development, the scenario where competitive pressure pushes labs to ship faster than they can verify safety.
Hassabis has also indicated that Artificial General Intelligence, the theoretical milestone where AI matches or exceeds human-level reasoning across domains, could emerge by 2029. That timeline makes the regulatory conversation feel less like a theoretical exercise and more like a countdown.
Here’s the thing: when the CEO of one of the world’s leading AI labs says the industry needs a referee, it carries a different kind of weight than when a senator says it. Hassabis isn’t an outsider lobbing criticism. He’s essentially asking for constraints on his own company’s operations.
The competitive chess game
There’s a less altruistic reading of Hassabis’s proposal that’s worth considering. Google DeepMind is one of the best-resourced AI labs on the planet. A regulatory regime that requires expensive pre-release testing and employs world-class technical reviewers would, by design, favor incumbents with deep pockets and established safety teams.
Startups and smaller competitors would face disproportionate compliance costs. Open-source AI development, which has been a major driver of innovation in the space, could be constrained if pre-release screening becomes mandatory.
Hassabis has advocated for global cooperation on AI safety, not just a US-centric approach. But by anchoring the proposal in American governance, he’s making a geopolitical bet: that the US is the only country with the institutional credibility, technical talent pool, and market leverage to make this work.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) ·