Employees at top AI labs urge US to back slowing of frontier development

Employees at the world’s largest AI companies have asked the U.S. government to help establish a framework to deliberately slow the pace of frontier model development so safety and governance tools can catch up.

The July letter, circulated under the heading “Pacing the Frontier,” garnered more than 1,100 employee signatures from labs including OpenAI, Anthropic, Google DeepMind and Meta, and calls on Washington to “support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development,” the organisers write.

1,178 signatories across OpenAI, Anthropic, Google and Meta

Organisers told reporters the petition attracted roughly 1,178 signatories; earlier tallies put the total at “more than 1,100,” reflecting rolling counts as employees added names. Named signatories cited in coverage include Anthropic CEO Dario Amodei, OpenAI figures such as Jakub Pachocki and Mark Chen, Meta AI scientist Shengjia Zhao, and Google safety lead Anca Dragan — signalling the petition spans senior researchers as well as rank-and-file engineers.

The initiative also has backing from two nonprofit groups, Guidelight AI Standards and Encode AI, which organisers say will help translate the demand for pacing into concrete technical and verification mechanisms.

Not a pause — a call for infrastructure to make "pacing" possible

The letter does not demand an immediate, blanket moratorium. Instead it asks governments and international partners to build the monitoring, auditing and governance tools that could enable a verifiable slowdown when required. “Support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development,” the petition reads in the language cited by reporters.

OpenAI publicly echoed that idea in summaries of its response, saying that “at some point in the future, AI acceleration for frontier model development may be so high that the world will need to pace the rate of AI advancement,” language that acknowledges both the risk and the difficulty of imposing a freeze.

That narrower ask matters because it tries to thread a political needle: governments, regulators and international bodies have resisted open-ended technology bans but could endorse standards and verification systems that permit conditional slowdowns.

Why employees say now — safety commitments, internal friction

Signatures spiked amid growing employee unease that safety practices have not kept pace with capability gains. A recent review reported some firms had weakened earlier safety commitments, a critique that organisers cite as part of the rationale for asking for external verification mechanisms.

The timing is also shaped by internal dynamics at the labs. Meta CEO Mark Zuckerberg told staff on July 2 that work on AI agents “had not accelerated in the way we expected,” a public admission that execution problems and shifting priorities are complicating firms’ trajectories and feeding anxieties inside engineering teams.

Skeptics point to those same frictions as proof that the “race” narrative is less simple than headline accounts suggest. Corporate management at times emphasises commercial timelines and product roadmaps; the employees’ petition, while broad, stops short of asking for an immediate industrywide halt — a sign signatories are trying to avoid a binary pause-or-go choice.

Critics outside the labs question whether governments can build credible verification fast enough. A study that flagged diluted safety pledges warned companies may be “moving the goalposts,” suggesting corporate promises could slip even with new governance frameworks in place.

Closing paragraph

The next test is whether Washington treats the petition as a call for diplomacy or a domestic regulatory agenda: expect lawmakers and agencies to weigh in in the coming months as they decide whether to fund or endorse the technical and audit infrastructure the letter requests. Observers will watch whether any international convening — or a concrete U.S. proposal — appears ahead of the autumn policy calendar.