Google DeepMind CEO Proposes New AI Regulatory Body for Frontier Models

Google DeepMind CEO Demis Hassabis is pushing for something that might sound contradictory in Silicon Valley: more oversight. In a Tuesday X post titled “A Framework for Frontier AI and the Dawning of a New Age,” Hassabis laid out a detailed proposal for a new regulatory body to oversee the release of frontier AI models. It’s a bold move that could reshape how the industry approaches safety and accountability.

The proposal centers on creating a standards body modeled after FINRA, the Financial Industry Regulatory Authority. Hassabis suggests that frontier AI labs would voluntarily submit their models for review up to 30 days before public release. The standards body would test them, identify risks, and develop best practices for deployment.

A Self-Regulatory Approach

What makes this proposal interesting is its structure. Rather than a heavy-handed government regulator, Hassabis envisions an independent organization backed by government but funded by the AI industry itself. This self-regulatory model draws inspiration from how financial markets operate. The team would include open source representatives, technical experts from AI companies, and specialists from growing AI safety organizations.

“The strength of this approach is it would be technically focused, while at the same time supporting innovation and incentivising responsible behaviour,” Hassabis argues. Once the assessment protocol proves itself effective, it could transition from voluntary to mandatory for any frontier model deployed in the US market.

This addresses a real problem. The U.S. government has already conducted ad hoc reviews on models like Anthropic’s Mythos and OpenAI’s Sol. Those reviews faced heavy criticism for lacking technical depth and making decisions behind closed doors. Nobody really knew how the government determined whether a model was safe enough to release.

The Political Reality

The timing of this proposal is particularly interesting given current political winds. White House AI advisor Sriram Krishnan recently said there “will not be an FDA for AI.” That statement reflects the Trump administration’s skepticism toward heavyweight regulation. By framing this as an industry self-regulatory organization rather than a government agency, Hassabis may have found a path forward that both tech companies and the current administration could accept.

The financial sector comparison is apt. Tech companies and policymakers understand how FINRA operates. It maintains market integrity while allowing innovation to flourish. Applying similar logic to AI could satisfy industry concerns about government overreach while addressing legitimate safety questions.

What Could Go Wrong

Of course, self-regulation carries its own risks. Will an industry-funded body actually say no to models that labs want to release? History suggests self-regulatory organizations can become captured by the industries they’re meant to oversee. The proposal would need strong governance structures and genuine independence to avoid becoming a rubber stamp.

Hassabis acknowledges this requires “ratcheting up” if risks intensify. The system would need to adapt as the field accelerates and new threats emerge. That’s harder to guarantee with a voluntary framework that depends on industry cooperation.

Still, this proposal represents genuine engagement with regulation rather than outright resistance. It suggests some AI leaders recognize that frontier models deserve closer scrutiny. Whether that sentiment is widespread enough to make this work remains unclear.

The stakes are high. Poor AI governance could create real harms. But overly restrictive regulation could stifle beneficial innovation. Finding that balance through something like Hassabis’ standards body might be more realistic than waiting for Congress to understand frontier AI well enough to legislate effectively.

Source: TechCrunch