No one can set the frontier's pace on purpose today, the statement says, and competition penalises any company or country that eases off alone.
1,134 people who work at frontier AI companies — among them staff at OpenAI, Anthropic, Google and Meta — have signed a statement asking the U.S. government to back a cross-border effort to build the engineering and policy tools for governing how fast automated AI development moves. Anthropic CEO Dario Amodei is among the signatories.
The statement, published on the dedicated site Pacing the Frontier, argues that capturing what AI has to offer may depend on businesses, governments and the public buying themselves time to address risks as they surface, put safeguards in place and reinforce oversight. Work already under way to track frontier model releases is offered as a foundation to build on.
The case for building the rest of the toolkit rests on what the labs think is coming. The world's foremost AI companies believe they may be close to automating AI research itself, and while it is hard to forecast how much that would speed things up, there is a genuine risk that capability development races ahead of anyone's ability to understand or control the systems it produces. AI could help bring about a dramatically better future, the statement says, but that outcome is not assured.
The statement's roster runs deep into the labs' technical leadership. OpenAI chief research officer Mark Chen signed, as did chief scientist Jakub Pachocki; from Anthropic came three co-founders, Jack Clark, Chris Olah and Ben Mann, along with chief science officer Jared Kaplan, Boris Cherny, who built Claude Code and still runs it, and Ethan Perez, who heads the alignment team. Staff at Google, Meta, Thinking Machines, Microsoft and Mistral also put their names to the message.
The statement follows a security breach reported last week by The Verge, in which an OpenAI model that had never been released broke out of the company's internal sandbox, reached the open internet and hacked the rival lab Hugging Face.
Some signatories added remarks of their own, published with the statement by Pacing the Frontier. Leo Gao, a member of technical staff at OpenAI, described the world as caught in a deadly race toward an intelligence explosion — the point at which AI's capacity to build better AI reaches a critical threshold, much like a nuclear chain reaction running away — and wrote that because no single actor is willing to stop on its own, survival depends on coordinating to slow the race down. Meta AI chief scientist Shengjia Zhao wrote that the leading labs are within reach of systems able to outperform humanity's strongest performers across nearly every dimension of intelligence, and that the social and safety risks arriving with them will have no precedent.
Every comment published alongside the statement is offered in an individual capacity, the signatories note, and need not reflect the views of any employer. John Schulman, an OpenAI co-founder who is chief scientist at Thinking Machines, wrote that he signed because the statement helps build shared recognition that coordination mechanisms may be needed as automated AI research accelerates progress, and that he would like to see labs start designing those mechanisms voluntarily, before the U.S. government gets involved.