DeepMind CEO Demis Hassabis has proposed creating an independent standards body, modeled after FINRA, to evaluate and regulate the release of frontier AI models. This move aims to ensure innovation while addressing the challenges of safe AI deployment.
- Proposed body to review AI models 30 days pre-release
- Modelled after Financial Industry Regulatory Authority (FINRA)
- Funded by AI industry but backed by U.S. government
What happened
Demis Hassabis, CEO of DeepMind, publicly advocated for the establishment of an independent standards body dedicated to regulating frontier AI models. The proposed organization would operate similarly to FINRA, providing voluntary early review of AI models by labs up to 30 days before release. Over time, this review could become mandatory for deployment in the U.S. market, ensuring that models meet safety and robustness criteria before public availability.
The new standards body would also collaborate with AI developers to remedy any critical issues discovered after a model’s release. This approach intends to replace the current ad hoc and sometimes criticized government-led reviews of AI models, offering a transparent, technically informed process to evaluate and manage AI risks effectively.
Why it matters
As AI models grow increasingly powerful and impactful, unregulated release poses safety, ethical, and security risks. The current regulatory ecosystem has faced criticism for lacking technical expertise and transparency, which undermines public trust and complicates responsible AI deployment. By establishing an independent, technically oriented standards body funded by the AI industry but supported by the government, DeepMind’s proposal aims to balance innovation incentives with stronger oversight.
This proposal also addresses political resistance to direct government regulation. By forming a self-regulatory organization with input from open source advocates, technical experts, and safety researchers, the standards body could adapt dynamically to emerging risks, promote best practices, and provide industry accountability without the need for traditional government agencies, which some policymakers continue to oppose.
What to watch next
Monitor reactions from key stakeholders including the U.S. government, AI labs, and safety organizations. The willingness of major AI developers to participate voluntarily could determine the effectiveness and credibility of the standards body. Additionally, legislative or executive responses that formalize or reject such a framework will be critical for shaping the future governance of advanced AI technologies.
Another factor to watch is the body’s capacity to integrate external safety groups and maintain independence while ensuring adequate funding from industry players. The operational details, staffing, and evaluation protocols will influence how well the organization can keep pace with rapid AI advancements and manage the highest-risk frontier models.