Demis Hassabis, CEO of Google DeepMind, has laid out a detailed proposal for overseeing the development of the most advanced artificial intelligence models. At its core is the creation of an independent US standards body, modeled after FINRA, the financial regulator, tasked with designing evaluation protocols for frontier AI systems before they are released to the public.
This new body would be responsible for establishing common safety tests and best practices across the industry, and would also have the authority to help coordinate a slowdown in AI development if serious risks were identified. Under the proposal, startups and research-only models would be exempt from this oversight, keeping scrutiny focused on large-scale commercial systems.
Hassabis framed the plan around what he described as a moment of deep technological uncertainty, noting that no one in the world can say with confidence what comes next for this technology. In his view, cautious optimism means building containment mechanisms now, before systems reach capabilities that become harder to monitor.
The proposal arrives as governments and tech companies actively debate how to balance innovation with public safety, with no unified regulatory framework yet in place in the United States. It remains unconfirmed whether the US government will formally adopt this model or how quickly such a structure could be implemented.
If realized, this body would mark a significant shift in how the most powerful AI models are released, adding a layer of external validation to a process that has so far relied largely on developers policing themselves.