Demis Hassabis, chief executive of Google DeepMind, on Tuesday published the most detailed regulatory blueprint yet from a leading AI executive, calling for a US-led standards body to test the most powerful artificial intelligence models before public release. The proposal, posted on X and Substack, lands weeks after the Trump administration ordered Anthropic to suspend foreign access to its Claude Fable 5 and Mythos 5 models, citing a jailbreak that breached national-security guardrails. That intervention, which forced the models offline for more than a fortnight, exposed the absence of structured pre-deployment oversight and has now catalysed a concrete industry counter-proposal.

The proposed agency would be modelled on the Financial Industry Regulatory Authority (FINRA), the private non-profit watchdog that polices Wall Street brokerages under federal supervision. It would be funded primarily by the AI industry, staffed with world-class technical experts and open-source representatives, and tasked with developing dynamic, regularly updated benchmarks that define “frontier-class” models. Developers would initially submit their most advanced systems for voluntary review up to 30 days before launch; once the testing regime proves reliable, compliance would become mandatory. The body would coordinate with federal agencies and US national laboratories to conduct security assessments, including stress tests for nuclear, biological and cyber risks.

Hassabis’s unilateralist approach diverges sharply from the international-coalition model advocated by OpenAI’s Sam Altman, who has urged a global treaty system. Viewed from Washington, the DeepMind proposal aligns with a White House instinct to leverage American market power: any foreign developer seeking access to the US market would need to meet the new standards, effectively exporting the framework. In European and Asian capitals, the idea is being read with caution. The G7 summit last month saw tech leaders press for unified safety rules, and the earlier US action against Anthropic stirred unease in Brussels and Tokyo. Hassabis’s essay, however, has already drawn public endorsements from Alphabet’s Sundar Pichai, Microsoft’s Satya Nadella and Mustafa Suleyman, OpenAI’s Altman, and Block’s Jack Dorsey—a signal that industry rivals are coalescing behind a US-centric regulatory architecture rather than waiting for a slower multilateral consensus.

Hassabis frames the urgency around his estimate that artificial general intelligence—systems matching the full range of human cognition—is “probably only a few short years away,” possibly by 2030. He describes the societal impact as potentially ten times the scale of the industrial revolution, unfolding at ten times the speed, and insists that philosophers, economists and broader civil society must help define the values embedded in that transition. The next factual milestone to watch is whether the US administration or Congress moves to establish such a body, or whether the proposal instead becomes a negotiating position ahead of the next G7 leaders’ summit, where AI governance is expected to feature prominently.