Hassabis Proposes FINRA-Like AI Regulatory Body, Draws Broad Industry Backing

Demis Hassabis, CEO of Google DeepMind, has published a proposal for a U.S.-led self-regulatory body modeled on FINRA to test frontier AI models for cyber, biological, and deception risks before public release [1][2][3]. Under the plan, labs would voluntarily submit models up to 30 days pre-release, with mandatory compliance intended once the body establishes credibility [1]. The proposal has drawn support across competitive and ideological lines, including from Microsoft, OpenAI, and investor Chamath Palihapitiya [4]. A parallel question is how an industry-led body would interact with government pre-release review already underway under a Trump AI executive order [5][6].

Frontier AI models currently face no mandatory independent pre-release review; labs largely self-certify. The Trump administration's improvised freeze of Anthropic's export-controlled models — producing two and a half weeks of ad-hoc negotiations with no established process — showed the cost of that gap [1]. If the proposal gains traction, it would be the first formal industry-wide pre-release safety mechanism, though its structural dependence on lab funding leaves the question of genuine independence open.

Demis Hassabis published a proposal for a national AI standards body modeled on FINRA, the Financial Industry Regulatory Authority that oversees brokerage firms in the U.S. [1][2][3]. The body would conduct independent pre-release testing of frontier AI models, with a testing scope covering cyber, biological, and deception risks. Labs would initially submit models voluntarily up to 30 days before release; Hassabis's design envisions mandatory compliance once the process proves effective, though the proposal does not specify what triggers that transition or who adjudicates it [1].

The proposal received early backing that cut across competitive and ideological lines. Microsoft and OpenAI, direct competitors of Google DeepMind, expressed support, as did Chamath Palihapitiya, an investor typically associated with the accelerationist camp in AI policy debates [4]. Hassabis won the 2024 Nobel Prize in chemistry for Google DeepMind's work on protein structure prediction and has operated largely outside the more contentious Silicon Valley debates; observers noted his distance from that arena as a factor in the proposal's reception as a credible focal point rather than a competitive move [4].

The immediate policy context includes overlapping government efforts. A Trump AI executive order already targets frontier model pre-release review [5], and a U.S. government agency has been separately tasked with safety testing frontier models before release [6]. The relationship between these government-led mechanisms and Hassabis's proposed industry body is unresolved — they could prove complementary, redundant, or in tension. The Trump administration's freeze of Anthropic's advanced models over export-control concerns, which produced two and a half weeks of improvised negotiations with no established rulebook, is cited by proponents as the kind of outcome a formal pre-release process would prevent [1].

Skeptics question whether an industry-funded self-regulatory body can actually constrain its own backers. The Neuron newsletter compared the structure to letting sports teams write their own officiating rules [1], and academic work on self-regulatory AI governance examines the same structural problem more formally [7]. Whether the broad initial support reflects genuine willingness to accept binding review — including adverse findings — or a softer form of endorsement is the central question the proposal leaves open.