The head of Google's AI lab is pushing for a US-led body to test frontier models before release, a proposal that aligns with White House momentum but carries an obvious tension at its center.
Hassabis wants an AI regulator modeled on FINRA, the Financial Industry Regulatory Authority, which oversees US broker-dealers using fees collected from the firms it regulates. Translated to AI, that structure would mean the companies racing to build the most advanced systems would also fund the organization tasked with determining whether those systems are safe to release. Hassabis argues industry funding keeps the body technically capable and independent from annual congressional budget fights. But the analogy cuts both ways. FINRA has faced persistent criticism that its industry ties blunt its enforcement instincts.
Hassabis made the proposal public around July 14, calling on the United States to establish an independent standards body that would test frontier AI models for national security and cybersecurity threats before commercial release. The body would also have authority to coordinate slowdowns across developers if serious dangers emerged. Sam Altman of OpenAI has made a similar call, which means two of the most prominent figures in frontier AI are now publicly asking for oversight of the work their own organizations are doing.
What the proposal actually asks for
The core mechanism is mandatory pre-release testing, a checkpoint that does not currently exist in any binding form in the United States. Under Hassabis's framework, no frontier model would reach the market without clearing an independent review focused on national security and cybersecurity risk. The body would set international standards as well, with the goal of establishing a governance framework before artificial general intelligence is achieved.
That last clause carries most of the weight. The proposal targets systems that do not yet exist rather than today's chatbots. It assumes that AGI, systems capable of matching or exceeding human performance across a wide range of cognitive tasks, is close enough that institutions need to be built now, before the capability arrives. Hassabis has framed the oversight body as a precautionary structure against AGI risks rather than a response to existing harms.
Get the latest model rankings, product launches, and evaluation insights delivered to your inbox.
The geopolitical frame
The insistence on US leadership is deliberate. Hassabis and others in this camp argue that fragmented national regulations, each country setting independent rules with no coordination mechanism, create gaps that bad actors can exploit and make it impossible to manage a global technology with global consequences. A US-led institution would set the baseline that other countries either adopt or negotiate against.
That argument has more traction now than it would have two years ago. The White House has shifted toward closer scrutiny of advanced AI work, and the political environment for some form of federal AI governance has grown materially more favorable since the EU AI Act came into force. The available sourcing does not specify whether US-led means the body is housed in Washington, staffed by American regulators, or anchored to American technical standards. That ambiguity would shape how other governments respond.
The enforcement gap
The proposal's most underdeveloped element is enforcement. Coordinating slowdowns across competing developers, including those outside the United States, requires either legal authority or extraordinary voluntary compliance. Hassabis has described neither in detail. A body that can test models but cannot force developers to delay release, or that has no jurisdiction over non-US labs, would function more as an auditor than a regulator.
That gap is not necessarily fatal to the idea. FINRA itself operates largely through the threat of expulsion from markets that firms cannot afford to lose access to. An AI standards body could operate on similar logic, where certification becomes a condition of operating in US markets. That gives it leverage without requiring direct legal authority over foreign entities. But the mechanism would need to be built into the proposal explicitly, and so far it has not been.
The White House's current posture toward advanced AI oversight makes the next several months a plausible window for this proposal to move from public argument to formal policy discussion. Whether the administration takes up the FINRA model specifically, or uses Hassabis's framing as one input among several, will become clearer as its AI governance agenda takes shape through the end of 2026.