Google DeepMind Chief Proposes FINRA-Style U.S. Oversight Body for AI

Google DeepMind CEO Demis Hassabis published a personal manifesto on Tuesday calling on the United States to create a new, independent oversight body for the most powerful AI models. The document, titled "A Framework for Frontier AI and the Dawning of a New Age," proposes an institution modeled on FINRA, the private financial markets regulator that operates under the oversight of the U.S. Securities and Exchange Commission.

Hassabis, a Nobel laureate and creator of the Gemini model, argues that the pace of AI development is outrunning our current understanding of the risks the technology poses. In the manifesto he writes that humanity has a "precious window" of time to build proper safeguards before the arrival of artificial general intelligence (AGI), which he describes as "just a few years" away.

A model borrowed from Wall Street

The centerpiece of the proposal is a reference to FINRA, the Financial Industry Regulatory Authority, a private regulator funded by the financial industry itself that oversees brokers and investment firms on Wall Street while remaining formally under the control of the U.S. Securities and Exchange Commission (SEC). Hassabis wants to bring that model to artificial intelligence: the new body would employ top technical experts, be funded by AI labs, but answer to Washington.

In the first phase, participation would be voluntary. Companies such as Google DeepMind, OpenAI, and Anthropic would share their newest models for safety testing up to 30 days before public launch. The tests would cover model capabilities in areas such as cyberattacks, biological weapons, and so-called "deceptive" behavior, meaning a model's ability to mislead its operators or users.

From voluntary to mandatory

Hassabis stresses that the voluntary stage is meant to be temporary. Once the testing system proves, in his words, "effective and robust," formalizing the rules should happen quickly, with frontier-class models required to pass testing before reaching the U.S. market. Crucially, the requirements would apply to all models in that class regardless of country of origin or whether they are released as open or closed systems, with benchmarks updated regularly as AI capabilities advance.

The proposal also calls for exempting startups and academic research from part of the requirements, to avoid a situation where costly compliance procedures end up favoring only large, well-funded companies like Google DeepMind itself. In an extreme scenario, the body would also have the power to coordinate an industry-wide slowdown in AI development if the risks began to outweigh the benefits.

Nobody in the world knows for certain what happens next - Demis Hassabis, CEO of Google DeepMind

What we collectively do now will decide how the next stage of civilization unfolds - Demis Hassabis, CEO of Google DeepMind

Why the US

Hassabis explicitly argues for U.S. leadership in this process, saying the country is "well positioned, given its economic and technical standing, to take the first step." The stance fits into a broader picture: the U.S. administration has spent months in talks with leading AI labs about rules for early access to new models, and Hassabis suggests he has spoken both with representatives of the Trump administration and with heads of other labs and European officials.

Hassabis himself stresses that the shape of any regulation should not be left solely in the hands of the people building the technology. In the manifesto he asks directly what values humanity wants to be guided by, what meaning and purpose will look like in a world reshaped by AI, and how the human condition itself might change, adding that these questions "cannot and should not be left to technologists alone."

What this means for the industry

Hassabis's proposal arrives as the question of global AI oversight keeps returning to the agenda, from a recently formed UN commission that includes the heads of Nvidia, Amazon, and Microsoft, to the U.S. administration's ongoing work on rules for early access to models. Unlike intergovernmental initiatives, the DeepMind chief's plan envisions a private body funded by the industry itself, which raises questions about how independent such oversight could really be from the interests of the largest labs, including Google's own.

So far, the manifesto has not drawn an official response from other AI labs or from the U.S. administration. Hassabis says, however, that discussions about the specific shape of such a body are already underway, and that his goal is to get the testing mechanism running before the end of 2026.

Sources: Axios (axios.com), The Decoder (the-decoder.com), Crypto Briefing (cryptobriefing.com), AOL (aol.com)