Google DeepMind CEO Demis Hassabis has put forward a regulatory framework to govern the release of frontier AI models. He wants a specialized standards body to replace the current patchwork of government reviews with a professional, technically rigorous oversight system.

The FINRA Model for Artificial Intelligence

In his proposal, “A Framework for Frontier AI and the Dawning of a New Age,” Hassabis models the new entity on the Financial Industry Regulatory Authority (FINRA). He envisions an independent organization, backed by the U.S. government but funded by the AI industry.

The mechanism would work in two phases. First, frontier labs would voluntarily hand their models to the Standards Body for a 30-day review before public release. After the assessment protocols prove effective, the system would shift to a mandatory requirement: frontier models would have to pass standardized tests to enter the U.S. market. This design catches critical vulnerabilities both before and after deployment.

Addressing the Flaws in Current Oversight

Hassabis pushes for a dedicated body because he sees today’s regulatory efforts as inadequate. He cites recent U.S. government reviews of Anthropic’s Mythos and OpenAI’s Sol as cautionary examples. Those assessments drew backlash for lacking deep technical expertise and for opaque decision-making about release timelines.

A self-regulatory organization would let the industry sidestep the political friction that now surrounds AI oversight. White House AI advisor Sriram Krishnan recently warned that an “FDA for AI” inside the executive branch is unlikely. A FINRA-style model offers a middle ground: technical independence without the bureaucratic hurdles of a direct government agency.

Technical Expertise and Scalable Safety

Hassabis stresses that the regulatory workforce must come from inside the industry and the open-source community. Staffed by people who understand model architecture and emerging risks, the body can evaluate models with real expertise.

The framework also allows a decentralized safety approach. The Standards Body could contract specialized AI safety groups for granular risk assessments. Because the organization would adapt to the field’s rapid acceleration, oversight could be “ratcheted up” dynamically as new, more severe risks appear. This keeps regulation a facilitator of responsible innovation rather than a bottleneck.

Key Takeaways

  • Hybrid Regulatory Model: Hassabis proposes an industry-funded, government-backed independent body modeled after FINRA to oversee frontier model releases.
  • Phased Implementation: The plan starts with voluntary 30-day pre-release reviews and moves to mandatory compliance for the U.S. market.
  • Technical Rigor: The body would draw on industry experts and open-source representatives to close the “expertise gap” seen in current government-led reviews.