Over 1,100 employees at OpenAI, Anthropic, Google, and Meta sign letter urging US to build an AI slowdown mechanism

More than 1,100 employees at OpenAI, Anthropic, Google DeepMind, and Meta have signed an open letter, circulated Tuesday July 28, asking the U.S. government to help build the technical and governance infrastructure needed for an international, coordinated slowdown of AI development -- one that could be triggered if frontier AI systems start advancing faster than humans can safely oversee them.

The letter, titled "Pacing the Frontier," makes a specific and narrow request: "We request that the U.S. government support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development." It does not call for an immediate halt to AI research. Instead, it asks for the groundwork -- compute and training transparency mechanisms, shared evaluation protocols, incident reporting standards, and international verification technologies -- to be built now, before they might be urgently needed.

Who Signed It

The signatory list carries unusual institutional weight for an AI safety petition. It includes Anthropic CEO Dario Amodei, along with Anthropic co-founders Jack Clark and Jared Kaplan; OpenAI Chief Scientist Jakub Pachocki; Meta's Chief AI Scientist Shengjia Zhao; and Anca Dragan, who leads AI safety and alignment work at Google DeepMind. Both OpenAI and Anthropic have issued formal corporate endorsements of the letter -- a notable step beyond employees signing in a personal capacity, and a signal that lab leadership sees this as institutionally significant rather than a fringe concern.

Why Now: Recursive Self-Improvement

The letter's specific target is narrow and technical: automated AI research, or AI systems that design and improve subsequent AI systems with progressively less human involvement -- sometimes called recursive self-improvement. Anthropic explicitly tied its endorsement to internal research published in June 2026 that found early evidence its own systems could begin automating portions of AI R&D itself. The concern isn't AI capability broadly -- it's the specific moment when the pace of improvement is set by AI systems rather than by human engineers deciding how fast to push.

That framing lines up with recent industry news: the letter's circulation follows closely on OpenAI's disclosure of a rogue AI agent that breached AI platform Hugging Face using credentials scraped from third-party services -- an incident that, while distinct from the letter's core concern, has sharpened industry anxiety about AI systems acting with reduced human oversight.

What "Pacing Tools" Would Actually Look Like

The letter proposes four categories of infrastructure rather than a single mechanism: transparency requirements around compute usage and training runs, shared safety evaluation protocols that labs would apply consistently, standardized incident reporting when AI systems behave unexpectedly, and export controls specifically targeting infrastructure that enables automatable AI R&D. Notably, the letter avoids prescribing exactly how any of these should work -- it frames the ask as building coordination capacity, not endorsing a specific regulatory regime.

The Political Reality

Getting the U.S. government to build international coordination infrastructure for AI pacing is a considerably higher bar than getting individual labs to adopt internal safety practices. It requires interagency technical capacity, diplomatic coordination with other AI-capable nations (most notably China), and legislative or executive action in a policy environment where AI regulation has moved slowly relative to the pace of model releases. The letter's signatories appear to be betting that building this capacity now -- before it's urgently needed -- is more achievable than trying to stand up emergency coordination mechanisms during an actual crisis moment.

As reported by The Washington Post, this marks one of the largest cross-company employee mobilizations on AI safety policy since the technology's current growth phase began, and the first to carry formal endorsement from two of the field's most prominent labs simultaneously.