AI
DeepMind CEO proposes independent regulator modeled on FINRA to oversee AI model releases
Demis Hassabis calls for a standards body to test frontier AI systems before deployment, drawing on financial industry precedent to balance innovation with safety oversight.
TechCrunch — AI
Google DeepMind's chief executive has outlined a vision for independent governance of cutting-edge AI systems, proposing a dedicated standards body that would evaluate frontier models before they reach the market. The framework, unveiled in a Tuesday social media post titled "A Framework for Frontier AI and the Dawning of a New Age," draws inspiration from how the financial industry regulates itself.
A New Regulatory Model#
Hassabis envisions a standards body structured after the Financial Industry Regulatory Authority (FINRA)—an organization that operates independently but serves a public function. Under this approach, the proposed regulator would initially ask AI labs to voluntarily submit frontier models for assessment up to 30 days before public release. Once this review process proves its effectiveness, the system could transition to mandatory compliance for models intended for deployment in the US market.
The body would also handle post-launch issues, working with labs to remediate critical vulnerabilities discovered after deployment.
Addressing Past Regulatory Gaps#
The proposal responds directly to criticism of existing government oversight. Earlier evaluations of Anthropic's Mythos and OpenAI's Sol models, conducted by US officials, faced significant backlash over limited technical expertise and opaque decision-making about when systems could be released. By transferring this responsibility to a specialized independent body, the framework aims to bring both rigor and transparency to the approval process.
Governance and Staffing#
The standards body would remain financially supported by AI companies but operate independently and free from direct government control—a structure that may help sidestep political resistance to AI regulation. The organization would draw expertise from open-source contributors, safety researchers, and technical specialists within the industry itself. Some evaluation work could be delegated to the expanding ecosystem of AI safety organizations, each focusing on distinct risk categories.
