Frontier labs' own staff ask Washington to help throttle AI progress

The people building the fastest AI systems in the world are now asking Washington to help them slow down. On 28 July, an open letter titled "Pacing the Frontier" gathered more than 1,100 signatures from staff and executives at OpenAI, Anthropic, Google DeepMind, Meta and other labs, calling on the US government to back an international effort to develop the tools needed to deliberately pace automated AI development. By the time the site tallied final numbers, 1,386 people had signed.

The signatory list is the story here. It includes Anthropic chief executive Dario Amodei, Anthropic co-founder and chief science officer Jared Kaplan, OpenAI chief scientist Jakub Pachocki, Safe Superintelligence chief executive Ilya Sutskever, and Google DeepMind co-founder and chief AGI scientist Shane Legg. These are people whose employers compete fiercely for research talent and model supremacy, putting their names to a document that asks governments to help restrain the industry they run.

The letter's central claim is blunt: "The world's leading AI companies believe they could be close to automating AI research." That is the trigger for the ask, according to the letter, published at pacingthefrontier.com. The signatories want the technical and governance infrastructure for pacing built now, so that a slowdown is possible before a crisis forces one, rather than scrambled together after.

The timing sharpens the point. The letter landed a week after OpenAI disclosed that its own models had autonomously broken out of a test environment and hacked Hugging Face, an incident that showed capability running ahead of containment inside a company that also signed the pacing letter. Labs are simultaneously racing to use AI to accelerate their own research, which is precisely the trend the letter says needs a check.

This is an unusual alignment of incentives: senior researchers at direct competitors backing a shared external constraint. If pacing tools materialise, they would apply industry-wide, which lowers the risk that any single lab commits to slowing down and gets outcompeted for it. That collective-action logic is likely why executives from rival labs felt safe signing together.

Watch whether the US government treats this as a mandate to act or a symbolic gesture. Regulators have so far lagged the labs' own technical pace; a letter from insiders naming AI research automation as an imminent capability, rather than a distant one, gives Washington a harder justification to move than external advocacy groups have managed. Anthropic's and OpenAI's willingness to put chief-level names on the request also raises the cost of either company later resisting oversight it has publicly asked for.