Demis Hassabis, CEO of Google DeepMind, is calling for a new independent organization to oversee the release of frontier AI models. In an X post published Tuesday, he outlined a proposal for what he calls a ‘standards body,’ modeled after the Financial Industry Regulatory Authority (FINRA), the self-regulatory organization that oversees U.S. broker-dealers.
The idea is straightforward: AI labs would voluntarily submit their most powerful models to this body for review up to 30 days before public release. The body would test them, flag risks, and develop best practices. Over time, that voluntary process could become mandatory, with frontier models required to pass review before they can be deployed in the U.S. market.
Hassabis frames the proposal as a way to build accountability without crushing innovation. ‘The strength of this approach is it would be technically focused, while at the same time supporting innovation and incentivising responsible behaviour,’ he writes, adding that it is ‘designed to keep up with the field’s acceleration and adapt to the biggest risks as they are identified.’
The proposal comes in direct response to recent government reviews of frontier models that went badly. The U.S. government conducted ad hoc reviews of Anthropic’s Mythos and OpenAI’s Sol, and both drew criticism for the same reasons:
- A lack of technical expertise among reviewers
- Opaque decision-making around when a model could be released
- No clear standards or benchmarks guiding the process
Hassabis wants to fix this by creating an organization that is backed by the U.S. government but funded by the AI industry and operated independently. Think of it as the industry paying for its own referee, with government legitimacy behind the calls that referee makes.
The political angle here matters. The Trump administration has shown little appetite for AI regulation. White House AI advisor and a16z general partner Sriram Krishnan recently said plainly that ‘there will not be an FDA for AI.’ By proposing a self-regulatory model like FINRA rather than a government agency, Hassabis is trying to thread that needle. A body that is industry-funded and technically staffed sidesteps the executive branch problem without leaving the field entirely ungoverned.
The proposed organization would be staffed by open source representatives and technical experts from inside the industry. It could also outsource specific evaluations to AI safety organizations that specialize in particular risk categories, which would tap into a growing ecosystem of independent safety researchers who currently have no formal role in model approvals.
Whether this gains traction is another question. AI regulation has been a third rail for much of the tech industry, and getting competing labs to agree on shared standards is historically difficult. But the fact that a sitting CEO of one of the most powerful AI research organizations in the world is publicly pushing for oversight is notable. It signals that at least some inside the industry believe the current situation, where the most capable models in history are released with minimal external review, is not sustainable.
