DeepMind CEO calls for an independent standards body to regulate frontier AI

In a recent post on X, Demis Hassabis, CEO of Google DeepMind, advocated for the establishment of an independent regulatory body tasked with overseeing the release of advanced AI models. His proposal, outlined in a document titled “A Framework for Frontier AI and the Dawning of a New Age,” suggests creating a standards organization similar to the Financial Industry Regulatory Authority (FINRA) that would evaluate these cutting-edge technologies and establish best practices for their deployment. Hassabis outlined a process where AI developers, referred to as Frontier Labs, would voluntarily submit their models for review to this new Standards Body up to 30 days prior to public release. He emphasized that once the assessment protocol is proven to be effective, it could quickly become mandatory, ensuring that all Frontier Models must pass this evaluation before entering the US market. Additionally, labs would collaborate with the Standards Body to mitigate any significant vulnerabilities that could arise after a model's release. This proposal aims to enhance the current system of ad hoc reviews conducted by the US government, which faced criticism for lacking technical expertise and transparency in decision-making regarding model releases. By creating a dedicated regulatory body, Hassabis believes that the responsibility for these assessments could be shifted to a new organization, which would be supported by the US government but operate independently with funding from the AI industry. The topic of AI regulation has sparked considerable debate within the tech sector and among policymakers, including members of the Trump Administration. Recently, White House AI advisor Sriram Krishnan dismissed the idea of an AI regulatory body within the executive branch, stating, “there will not be an FDA for AI.” However, Hassabis proposes that establishing a self-regulatory organization akin to FINRA could alleviate some of these concerns. He envisions this new regulator being staffed by technical experts from the industry as well as representatives from the open-source community, ensuring a diverse range of insights. Financial support from AI labs would be crucial to attract and retain talented personnel. Furthermore, the regulator could enlist specialized AI safety groups to handle specific risks effectively. Hassabis argues that this approach would not only be technically focused but also foster innovation while encouraging responsible practices. He believes it is essential for the regulatory framework to keep pace with the rapid evolution of AI, adapting to emerging risks as they are identified, and enhancing its rigor as necessary depending on the severity of the situation.