Google DeepMind CEO Demis Hassabis is calling for a new independent body to test and certify frontier AI models before they ship in the US, modeled on the Financial Industry Regulatory Authority. In an X post Tuesday morning titled "A Framework for Frontier AI and the Dawning of a New Age," Hassabis proposed that labs voluntarily submit models to the group up to 30 days before release, with mandatory compliance to follow once the process is proven. It is the most concrete regulatory blueprint yet from a sitting frontier-lab CEO.
The proposed standards body would be backed by the US government but funded by the AI industry itself and operated independently — the same self-regulatory structure that governs broker-dealers through FINRA. Hassabis argues that structure sidesteps the objections to a federal AI agency while still giving reviews teeth. Under the framework, once the assessment protocol was shown to be effective, passing it would become a precondition for deploying frontier models in the US market.
“Initially, Frontier Labs would voluntarily share models with the Standards Body for review up to 30 days before release”— Demis Hassabis, CEO of Google DeepMind
The proposal explicitly builds on the ad hoc reviews the US government has already conducted on Anthropic's Mythos and OpenAI's Sol. Those reviews drew sharp criticism for a lack of technical expertise inside government and opaque decision-making about when a model was cleared for release. Handing the work to a purpose-built organization staffed by technical experts is Hassabis's answer to both complaints.
Key facts
- 01Hassabis proposes an independent standards body modeled on FINRA, backed by the US government but funded by AI labs.
- 02Frontier labs would voluntarily submit models for review up to 30 days before release, with mandatory compliance to follow.
- 03The framework builds on ad hoc US government reviews of Anthropic's Mythos and OpenAI's Sol, both criticized for opaque decisions.
- 04White House AI advisor Sriram Krishnan has said 'there will not be an FDA for AI,' complicating any executive-branch regulator.
- 05The body would be staffed by open-source representatives and technical experts, with evaluations potentially outsourced to AI safety groups.
Staffing is where the FINRA analogy does the most work. Hassabis envisions the body drawing from open-source representatives and technical experts inside the industry, with financial backing from labs that would be required to retain them. Some evaluations could be outsourced to the growing pool of AI safety groups that specialize in specific risks — biosecurity, cyber, autonomy — rather than concentrating all expertise in one office.
The political obstacles are steep. AI regulation remains contested inside both the tech industry and the Trump Administration, and White House AI advisor Sriram Krishnan, a general partner at a16z, recently dismissed the idea of an executive-branch AI regulator.
“there will not be an FDA for AI.”— Sriram Krishnan, White House AI advisor and a16z general partner
A self-regulatory organization is a deliberate workaround to that constraint. FINRA is not a federal agency; it is an industry-funded nonprofit that the SEC oversees. Structuring an AI standards body the same way lets the government stay one step removed while still granting the reviews legal weight, which is roughly the compromise Hassabis is proposing.
Hassabis frames the design as one that keeps pace with the technology rather than freezing it. He argues the body could ratchet up scrutiny as risks are identified and adapt to new capabilities without waiting on legislation.
“The strength of this approach is it would be technically focused, while at the same time supporting innovation and incentivising responsible behaviour”— Demis Hassabis, CEO of Google DeepMind
The unresolved question is whether the other frontier labs would sign on. Voluntary pre-release review means handing over model weights, evaluation results, and capability data to an outside organization potentially staffed with representatives from competitors — a bar that OpenAI, Anthropic, xAI, and Meta have so far cleared only on their own terms. Hassabis's proposal offers no enforcement mechanism during the voluntary phase, and skeptics inside the labs are likely to note that a standards body funded by its regulated entities carries obvious capture risk, the same critique long leveled at FINRA itself.
The proposal lands amid a broader push from AI leaders warning that self-governance is not sustainable indefinitely. It follows Hassabis's earlier call, covered here last week, for a US-led global AI watchdog with authority to halt frontier deployments — the standards body reads as the domestic-first version of that vision, structured to be politically achievable.
For the AI industry, a FINRA-style body would be the least intrusive regulator on offer: technically competent, industry-funded, and free of the drug-trial-style delays that a true FDA-for-AI would impose. That is precisely why it may be the version that actually happens. If DeepMind, Anthropic, and OpenAI can align on a shared review protocol, the White House gets a regulatory answer without owning a new agency, and the labs get predictability on release timelines — a tradeoff that looks increasingly attractive as capability gaps between models narrow and the political cost of the next high-profile incident keeps rising.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
