Google DeepMind CEO Demis Hassabis has called for the creation of an independent regulatory body to review and test advanced frontier AI models before they are released.
In an X post titled “A Framework for Frontier AI and the Dawning of a New Age,” Hassabis proposed establishing a technical standards body modelled on the Financial Industry Regulatory Authority, or FINRA.
The proposed organisation would assess powerful AI models, develop best practices for their release, and work with AI laboratories to address critical vulnerabilities discovered after deployment.
How Would the Proposed AI Regulator Work?
Under Hassabis’ framework, frontier AI laboratories would initially submit their models voluntarily for review up to 30 days before release.
Technical experts would use that period to evaluate risks related to safety, security, misuse, and other potentially harmful capabilities.
Once the assessment process proves effective and reliable, the system could be formalised. Frontier AI models would then be required to pass the review before being deployed in the US market.
AI companies would also be expected to cooperate with the standards body in resolving serious vulnerabilities identified after a model’s release.
A FINRA-Style Body for Frontier AI
Hassabis suggested modelling the organization on FINRA, a self-regulatory body that oversees parts of the US financial industry.
Rather than operating as a traditional government agency, the AI regulator could be supported by the US government, funded by the AI industry, and managed independently.
This structure could help address concerns that government agencies may lack the technical expertise and flexibility required to keep pace with rapidly advancing AI models.
The body would focus primarily on technical evaluations rather than broader political or commercial decisions.
Proposal Follows Criticism of Previous AI Reviews
The proposed framework builds on earlier US government reviews of advanced models, including Anthropic’s Mythos and OpenAI’s Sol.
Those reviews faced criticism over limited transparency, insufficient technical expertise, and unclear decision-making about whether a model should be approved for release.
Hassabis’ proposed regulator would move those assessments to a specialised organisation staffed by experts capable of conducting more consistent and detailed evaluations.
This could create a clearer process for reviewing frontier models before they reach the market.
Who Would Staff the AI Standards Body?
Hassabis envisions the organisation bringing together technical experts from the AI industry, representatives from the open-source community, and independent researchers.
Funding from major AI laboratories could allow the body to recruit specialists in model evaluation, cybersecurity, misuse prevention, and AI safety.
The organisation could also outsource certain evaluations to independent safety groups with expertise in specific risk areas.
Separate teams could assess biological risks, cybersecurity capabilities, autonomous behaviour, deception, and the potential use of AI models in harmful activities.
AI Regulation Remains Controversial
The proposal comes as AI regulation remains a divisive issue across the technology industry and the US government.
Supporters of stronger oversight argue that frontier models could create serious economic, security, and social risks that require independent testing.
Critics warn that strict regulation could slow innovation, strengthen the position of major technology companies, and make it harder for smaller laboratories and open-source developers to compete.
White House AI adviser and Andreessen Horowitz general partner Sriram Krishnan previously rejected the idea of creating a traditional federal regulator similar to the US Food and Drug Administration, saying there would not be an “FDA for AI.”
A FINRA-style self-regulatory organisation could offer a compromise by providing formal oversight without placing the entire process within the executive branch.
Balancing Innovation and AI Safety
Hassabis argued that the main strength of his proposal is its technical focus.
The standards body would aim to promote responsible behaviour while allowing companies to continue developing and releasing new AI systems.
Its assessment methods could evolve as frontier models become more capable and new risks emerge.
Requirements could also be strengthened if future developments show that advanced AI systems pose more serious threats than previously expected.
Could Frontier AI Reviews Become Mandatory?
The system would begin on a voluntary basis, giving AI companies time to test and improve the assessment process.
If the framework proves robust, the US government could formalise it and require frontier models to receive approval before entering the American market.
Such a move would represent a major change for the AI industry, where companies currently conduct many safety evaluations internally and publish varying levels of information about their testing methods.
Mandatory independent reviews could create common standards and reduce the likelihood of powerful models being released without adequate safety checks.
A New Approach to Frontier AI Regulation
Hassabis’ proposal reflects growing pressure to create oversight mechanisms that can keep pace with the rapid development of artificial intelligence.
A regulator backed by the government, funded by the industry, and operated independently could provide deeper technical expertise than a traditional agency while reducing concerns about direct government control.
However, questions would remain about the body’s independence, funding model, enforcement powers, transparency, and treatment of smaller AI developers.
The debate is likely to intensify as frontier models become more powerful and governments reconsider whether voluntary safety commitments are sufficient.
Read the article in
