DeepMind CEO's Urgent Plea: Independent Body Needed to Tame Frontier AI

Google DeepMind CEO Demis Hassabis has proposed the establishment of a new regulatory body to oversee the release of frontier AI models. In an X post titled “A Framework for Frontier AI and the Dawning of a New Age,” Hassabis advocated for a “standards body” structured similarly to the Financial Industry Regulatory Authority (FINRA). This proposed organization would be responsible for rigorously testing frontier models and developing comprehensive best practices for their deployment.

Initially, Hassabis envisions a voluntary system where Frontier Labs would submit their models to the Standards Body for review up to 30 days before public release. Once the assessment protocols prove effective and robust, the framework could quickly transition to a formal, mandatory system. Under such a system, Frontier Models would be required to pass the body's evaluations before being deployed in the U.S. market. Additionally, labs would collaborate with the Standards Body to promptly address any critical vulnerabilities identified post-release, ensuring ongoing safety and stability.

This proposed system aims to improve upon previous ad hoc reviews conducted by the U.S. government on models like Anthropic’s Mythos and OpenAI’s Sol. Those earlier reviews faced significant criticism due to a perceived lack of technical expertise and opaque decision-making processes regarding model releases. Hassabis's vision suggests that these critical decisions could be delegated to a new, independent organization, backed by the U.S. government but primarily funded by the AI industry itself.

The broader prospect of AI regulation remains a contentious issue within both the tech industry and political spheres, including the Trump administration. Recently, White House AI advisor Sriram Krishnan notably dismissed the idea of an AI regulator within the executive branch, stating, “there will not be an FDA for AI.” Establishing the standards body as a self-regulatory organization, much like FINRA, could serve as a pragmatic approach to alleviate some of these industry and governmental concerns.

Hassabis further elaborates on the structure of this regulator, imagining it staffed by a diverse group of experts, including representatives from open-source communities and technical specialists from within the AI industry. The necessary financial backing from AI labs would ensure the retention of top talent. The body could also strategically outsource certain evaluations to the burgeoning ecosystem of AI safety groups, allowing them to specialize in specific risk domains. Hassabis asserts that the strength of this approach lies in its technical focus, its capacity to support innovation, and its ability to incentivize responsible behavior, while also being adaptable to the field's rapid advancements and capable of scaling its efforts as the seriousness of identified risks demands.