WASHINGTON: The head of Google's artificial intelligence (AI) lab called on July 14 for the United States to create an organisation that would test the most powerful AI systems before they can be released to the public.
Demis Hassabis, who runs Google DeepMind and won a Nobel Prize for his work on protein science, said in an essay that AI capable of matching the full range of human thinking, known as artificial general intelligence, is likely just a few years away.
Hassabis leads the division behind Google's Gemini family of AI models, which compete directly with offerings from OpenAI and Anthropic in the fast-moving race to build more powerful systems.
The proposal lands amid growing calls from rival AI executives for binding government oversight of the technology – and weeks after a bruising standoff between the Trump administration and Anthropic that laid bare the absence of clear rules for the sector.
In June, the Commerce Department ordered Anthropic to suspend all foreign access to its Claude Fable 5 and Mythos 5 models, citing national security concerns linked to a reported jailbreak – users successfully blowing past guardrails designed to contain them.
The company was forced to take both models offline entirely for more than two weeks before access was gradually restored.
Hassabis proposed a new regulatory body modelled on the Financial Industry Regulatory Authority (FINRA), the private watchdog that polices Wall Street brokerage houses under federal oversight.
He said it would be staffed by leading technical experts, funded primarily by the AI industry, and tasked with developing safety benchmarks that frontier models must clear before entering the market.
Under his plan, AI developers would initially submit their most powerful models for review on a voluntary basis up to 30 days before release. Once the testing regime proves reliable, compliance would become mandatory.
"Nobody in the world knows for sure what is going to happen from here, and even the experts disagree," Hassabis wrote.
"When there is a large degree of uncertainty and the stakes are this high, proceeding with cautious optimism is the sensible and correct strategy."
The rules enforced by the new US agency would apply to any model deemed powerful enough to qualify as "frontier-class" – whether built in the US or abroad, with Hassabis hoping that international players would buy into the system once it was up and running.
In an interview with Axios published alongside his manifesto, Hassabis described the Mythos episode as "a bit of a wake-up call" that showed Washington needs a more structured approach.
Anthropic CEO Dario Amodei has issued his own call for regulation days before the government crackdown, publishing an essay in early June that advocated for a federal agency with the power to vet powerful AI models before deployment. – AFP
