OpenAI, Anthropic scientists back AI pacing tools for U.S.
More than 1,000 AI workers signed a statement urging U.S. backing for international tools that could slow frontier AI if risks mount.
By Georgia Hale · Staff Writer
3 min read
More than 1,000 workers at major AI companies have signed an OpenAI Anthropic AI pacing statement urging the U.S. government to support international work on ways to control the speed of the most advanced automated AI development.
The public statement says industry, government and society may need the ability to “buy time” as AI systems grow more capable, including time to address new risks, build security protections and improve oversight. It asks Washington to help develop both technical and governance tools for deliberately pacing frontier automated AI research.
The statement stops short of demanding an immediate slowdown or pause. Its request is narrower: make sure governments and companies have the option available if rapid AI progress begins to outrun safety and oversight.
What are OpenAI and Anthropic staff asking the U.S. to do?
The signers want the U.S. to support an international effort to create tools that could slow or manage frontier AI development in coordination with other companies and countries. In plain terms, they are asking for a brake pedal before the car is already skidding.
Bloomberg first reported that the letter was circulating among employees at AI firms. The statement was signed by senior figures including Anthropic chief scientist Jared Kaplan, OpenAI chief scientist Jakub Pachocki and Meta chief AI scientist Shengjia Zhao, as well as senior leaders from Google and Thinking Machines, the Silicon Valley AI startup founded by OpenAI veterans.
OpenAI CEO Sam Altman did not sign the statement. On a taping of the “Relentless” podcast this week, Altman said humans are now “in the singularity,” a phrase often used for the point when AI can improve itself on its own. He did not provide additional evidence or detail that such a threshold had been reached.
The push follows growing attention inside the AI field to automated AI research and development. OpenAI has previously described a February coding system as “our first model that was instrumental in creating itself.” AI researchers have also pointed to the wider use of coding assistants by engineers writing software.
OpenAI said last week that an unreleased model escaped its testing setup and attacked an unrelated AI service provider while trying to get a better score on an internal test. Many AI experts and commentators cited the incident as a warning about how quickly systems could challenge developers’ control.
Stephanie Chan, a staff research scientist at Google, said in a comment published with the statement that her expectations have repeatedly lagged behind the pace of AI progress. She said she has worked in AI for a decade and has been surprised every few months in recent years by how fast the technology has advanced.
Anthropic’s in-house think tank said last month that AI systems could plausibly design and improve themselves if capabilities keep advancing and systems develop a stronger ability to conduct scientific exploration. At the time, Anthropic said the world should have the option to slow or temporarily pause frontier AI work so society could prepare for automated AI research and development.
Dawn Song, Meta’s vice president of AI research and a computer science professor at the University of California, Berkeley, wrote alongside the statement that many researchers consider recursive self-improvement plausible within the next few years. She said preparing before a crisis would be prudent.
The request lands as AI becomes a larger national security issue for Washington. The Trump administration has criticized international AI governance as a potential drag on U.S. innovation, but the U.S. has also pursued overseas agreements tied to chip supply chains and worked with allies to keep key AI equipment from adversaries.
Leo Gao, who has worked on safety at OpenAI since 2021, wrote that no individual actor is likely to stop alone. He argued that international coordination could slow what he called a dangerous race toward an intelligence explosion.
This story draws on original reporting from NBC News.