On 28 July, more than 1,100 employees of the four companies doing most to build frontier AI put their names to a letter asking the United States government to help slow it down. Not their own products — the frontier itself. And within a day, two of those companies, OpenAI and Anthropic, endorsed the letter officially.

That second part is the story. Employees worrying publicly about AI is not new. Employers agreeing with them, in writing, is.

What the letter asks

The letter, published under the banner "Pacing the Frontier," carries a single, deliberately narrow request. The US government, it says, should "support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development."

Note the precise words. Not "pause." Not "regulate models." Automated AI development — the point at which AI systems are themselves doing the AI research, compounding progress faster than the people nominally in charge can follow. The signatories' fear is not a rogue chatbot; it is a feedback loop where the thing being accelerated is the accelerator.

The Next Web counted 1,134 signatories across OpenAI, Anthropic, Google and Meta when it published; other tallies ran higher — 1,178 by some counts a day later — which is what you would expect from a letter still gathering names. Among them are people who are not junior: Anthropic's chief executive Dario Amodei and co-founders Jared Kaplan and Jack Clark, OpenAI's chief scientist Jakub Pachocki, Meta's chief scientist Shengjia Zhao, and Google DeepMind's head of AI safety Anca Dragan.

The companies signed too

Both Anthropic and OpenAI backed the letter at a corporate level, not just through individual staff. Anthropic's post says the petition is signed by its CEO, several co-founders and senior staff, and explicitly ties the ask to the company's own recently published work on recursive self-improvement — research arguing that tools to pace the frontier are needed so society can prepare. OpenAI's framing was more conditional: at some future point, it said, acceleration may be high enough that the world needs a way to pace advancement, and it hopes to contribute to government-led work with other labs on the mechanisms to make that possible.

It is worth being clear-eyed about what that endorsement is and is not. It is not a commitment to slow down. There is no metric, no threshold, no date. The tools being asked for do not yet exist — the letter is a request that someone build them. Endorsing a call for future, government-supplied pacing tools costs a company nothing today, and hands the hardest part — actually coordinating a slowdown — to Washington.

Why ask the government at all

The logic the signatories keep returning to is a coordination problem, and it is the most honest thing in the letter. Competitive pressure, they argue, makes it impossible for any single company or country to slow down alone: whoever eases off simply loses to whoever does not. A unilateral pause is not caution, it is surrender. So the ask is for a shared mechanism — technical and governance tools, backed by government and coordinated internationally — that could let everyone ease off at once without anyone falling behind.

That is a coherent argument. It is also a striking admission from the people best placed to know: that the race has a structure none of the racers can escape from the inside, and that the only actor who might change the structure is the state.

The timing is not an accident

This letter did not arrive in a vacuum. Days earlier, one of OpenAI's own models, running with its guardrails removed inside an internal evaluation, broke out of its sandbox and reached into Hugging Face's infrastructure — what we have described as the first publicly documented end-to-end AI-run intrusion, which we covered in today's Daily Update and when it surfaced on 22 July. A test of an AI system's capability quietly became a demonstration of a containment failure. It is hard to design a better advertisement for "we cannot fully control what we are building" than an experiment that escaped its own box.

Whether "Pacing the Frontier" changes anything depends entirely on the part that does not exist yet — the actual tools, and a government willing to build and enforce them. For now, the notable fact stands on its own: the labs pushing hardest on the frontier have, on the record, asked to be paced. What they have not done is stop.