Artificial intelligence company Anthropic has called for a worldwide slowdown in the development of the most powerful AI systems, warning that the latest models are beginning to show signs they could escape human control. The San Francisco-based company, which makes the Claude family of AI models, published a report on Thursday arguing that a global pause on cutting-edge AI development would "likely be a good thing" — but cautioned that any unilateral halt by a single company would simply allow rivals to race ahead.

For such a pause to be meaningful, Anthropic said, it would require multiple major AI companies across multiple countries — most notably the United States and China — to agree simultaneously to stop, under a set of rules that all parties could independently verify. "Without a global coordination mechanism, companies and governments will have to make difficult decisions about safety while under competitive and geopolitical pressures," the company said. Anthropic drew a comparison to nuclear arms control treaties, while acknowledging the challenge would be even greater: AI training is far easier to conceal than a missile silo, and the incentive to quietly continue development would be enormous.

The proposal faces significant resistance in Washington and Silicon Valley, where officials and tech executives have long argued that any slowdown risks handing China a decisive strategic advantage in what many regard as the defining technology race of the century. Critics within the industry and at the White House have also accused Anthropic of overstating worst-case risks as a way of slowing competitors under the guise of safety. The company's own Mythos model — a system the White House has acknowledged as particularly powerful — has not been released to the general public due to its cybersecurity capabilities and is currently available only to a small number of vetted organisations.

Some signals from Washington suggest a degree of openness to engagement on the issue. President Donald Trump said he discussed the possibility of US-China cooperation on AI safety during a recent visit to Beijing, and this week signed an executive order giving the government a 30-day window to conduct preliminary reviews of the most powerful American AI models before their public release.

Underpinning Anthropic's call is internal data showing that AI is already dramatically accelerating its own development — a feedback loop the company warned could eventually lead to "recursive self-improvement," in which an AI system becomes capable of making itself smarter with little human input. "We are not there yet, and recursive self-improvement is not inevitable," the report said, adding that it could arrive sooner than most governments and institutions are prepared for. "The evidence suggests that the human role is narrowing at each step in the AI development process," the company said. Anthropic announced plans to convene government officials, scientists, advocacy groups and competing AI firms in the coming months to explore how a workable coordination framework might be designed.