More than 1,100 employees from OpenAI, Anthropic, Google, Meta, NVIDIA and other frontier
artificial intelligence (AI) companies have signed a statement urging the US government to support an international effort to develop the technical and governance tools needed to "deliberately pace the frontier of automated AI development". The signatories include OpenAI chief scientist Jakub Pachocki, Anthropic CEO Dario Amodei, Meta AI chief scientist Shengjia Zhao and senior researchers across leading AI labs.
The statement is not a call to pause AI development. Nor does it propose slowing today's models. Instead, it argues that governments should build mechanisms that would allow society to "buy time" if frontier AI begins improving itself faster than existing institutions can understand, regulate or control.
Why researchers are worried now
The concern is centred on automated AI research, sometimes referred to as recursive self-improvement. Today's AI systems already assist researchers by writing software, analysing experiments and helping develop future AI models. While humans remain firmly in control, frontier labs increasingly rely on AI to perform parts of the research and engineering process. If AI systems eventually become capable of improving the next generation of AI with minimal human involvement, capability gains could compound much faster than they do today.
The petition reflects growing concern inside frontier labs that such a feedback loop may be approaching sooner than previously expected.
"It is hard to predict exactly how much this will accelerate AI progress," the statement says, "but there is a real risk that capability development rapidly accelerates beyond our ability to understand or control the resulting systems." It adds that industry, governments and society may need the option to "buy time" to address emerging risks, strengthen security measures and improve oversight.
That concern closely mirrors arguments made by Anthropic in recent months. The company has previously said the world should have the option to temporarily slow frontier AI development if recursive self-improvement significantly accelerates progress, a position it reiterated while supporting the latest initiative.
Why companies cannot slow down on their own
One of the central arguments made by the signatories is that individual companies cannot realistically reduce the pace of development by themselves.
Every frontier AI developer is competing against rivals with similar ambitions, while governments increasingly view AI leadership as a matter of national competitiveness. Under those conditions, voluntarily slowing development risks simply allowing competitors to move ahead.
Leo Gao, a member of the technical staff at OpenAI, described the problem in starker terms. "The world is locked in a deadly race towards an intelligence explosion," he wrote. "Going slower would give us much-needed time to make it go well, but no individual actor is willing to stop unilaterally. To survive, we must coordinate to slow down the race."
Buying time, not stopping progress
The signatories are not arguing that AI development should stop. Rather, they argue governments should have technical and governance tools ready before they become necessary.
Several researchers said the pace of AI progress itself has repeatedly exceeded their expectations.
Stephanie Chan, a staff research scientist at Google, wrote that despite working in AI for a decade, she continues to be surprised by how rapidly the technology advances. She warned that AI may end up progressing far more quickly than societies and governments can adapt, making it important to prepare mechanisms that could deliberately slow frontier development if required.
Meta's vice president of AI research, Dawn Song, framed the issue more broadly.
"If it's true that AI will be as transformational as electricity, we better be intentional about how it is introduced to the world," she wrote. "A mindless race without assurances on governance, safety or the technology serving the public interest stands in the way of such intentionality."
Anthropic researcher Ethan Perez also warned that safety teams are already struggling to keep pace with rapidly evolving capabilities.
"With the current rate of AI progress, safety teams at AI companies have to sprint to prevent new risks to society every few months. At some point, we're going to hit problems we need more time to solve," he wrote.
Companies publicly back the petition
The initiative has also received public support from companies most closely associated with frontier AI. In a post on X, OpenAI said it believes there could come a point where "AI acceleration for frontier model development may be so high that the world will need to pace the rate of AI advancement." The company added that it hopes to contribute alongside the US government, other AI labs and the open-source community in developing the tools that could make such coordination possible.
Anthropic also endorsed the petition on X, saying its recent research on recursive self-improvement points to the need for tools that can deliberately pace frontier AI development so society has time to prepare. The company said it was encouraged to see broad agreement emerging across the AI field.
The proposal does not outline what those governance mechanisms should ultimately look like. Some signatories also cautioned that any future system should avoid becoming an unnecessary regulatory barrier or favouring a handful of large AI companies over smaller competitors.
John Schulman, chief scientist at Thinking Machines and a former OpenAI researcher, argued that companies should begin preparing even before governments step in.
"I'd also like to see labs start designing these mechanisms voluntarily, even before the USG gets involved," he wrote.
Whether governments act on the proposal remains uncertain. But the petition highlights an important shift inside frontier AI labs. The people pushing the technology forward are certain that AI will continue advancing rapidly, however, their concern is whether society will have a credible way to slow that progress if it ever begins moving faster than humans can safely manage.
 "Why over 1000 AI researchers want governments ready to slow AI progress")
