'A precious window' before AGI: DeepMind's Demis Hassabis proposes US-led AI watchdog

Google DeepMind CEO and co-founder Demis Hassabis has raised the alarm over how close highly powerful artificial intelligence is and the sweeping changes it could bring. His latest warning underscores the need for prompt action to prepare for the arrival of artificial general intelligence (AGI).

In a short essay published on X and Substack on Tuesday, the tech executive argued that there is a "precious window" during which governments and industry leaders can act before AGI arrives. AGI refers to AI systems capable of performing cognitive tasks at or near human levels.

Through the essay, Hassabis proposed a US-led coalition that would review the world's most advanced AI models to ensure they are safe before being released to the public. His comments come amid a series of recent interventions by the Trump administration aimed at limiting the release of advanced AI models from companies such as OpenAI and Anthropic.

More from Tech

Hassabis has been raising concerns about advanced AI for years and has repeatedly stressed the need for a formal body of experts to oversee the development of frontier AI models.

Hassabis builds support for proposal

According to a report by Axios, Hassabis has been quietly building support for the proposal, including briefing officials in the Trump administration, executives at leading AI labs, and European policymakers. He hopes to have the organisation up and running before the end of the year. Hassabis also told Axios that the proposal has received a positive response from the Trump administration.

The proposal marks one of the latest efforts by Hassabis and other industry leaders to establish a coherent framework for governing increasingly powerful AI systems while mitigating the risks they may pose.

At present, there is no comprehensive global framework specifically governing AI. As part of his broader push for stronger AI oversight, Hassabis last month signed a statement calling for tougher safeguards against AI-assisted bioweapon development.

Hassabis' latest proposal and public remarks come shortly after leading economists and technology executives, including Anthropic co-founder Jack Clark and former Google CEO Eric Schmidt, urged world leaders to take the looming economic impact of AI more seriously.

The standards body to be modelled on FINRA

Hassabis is proposing an AI standards body modelled on FINRA (Financial Industry Regulatory Authority), a privately funded industry watchdog that oversees Wall Street under the supervision of the US Securities and Exchange Commission (SEC). According to a report by Axios, frontier AI labs would initially submit their models to the body voluntarily, up to 30 days before release, for safety testing aimed at identifying dangerous cyber, biological, and deceptive capabilities.

Once the testing framework proves to be effective and robust, it could eventually be formalised. Under such a system, frontier AI models would be required to pass the safety assessments before they could be deployed in the United States.

Hassabis envisions the board comprising Turing Award winners and other credentialed experts, alongside representatives from industry, government, and the open-source community.

Anthropic CEO Dario Amodei has also warned about the risks posed by increasingly powerful AI systems and has called for binding regulation. He has proposed an FAA-style regulatory agency with the authority to block the release of AI models deemed unsafe.