On July 28, a website went live at a plain address: pacingthefrontier.com.
On it sat a single sentence, and a list of names. The sentence asked the US government to support an international effort to build the technical and governance tools needed to deliberately slow the frontier of automated AI development.
The names are what make it matter. As of this morning the counter had passed twelve hundred and was still climbing. Among the signatories: Dario Amodei, the CEO of Anthropic. Ilya Sutskever, who co-founded OpenAI. Jakub Pachocki, OpenAI's chief scientist. Jared Kaplan and Chris Olah, Anthropic co-founders. The chief scientists of the biggest labs on earth, signing a letter that asks someone to help slow their own field down.
Within hours, OpenAI and Anthropic endorsed it at the company level. The two fiercest rivals in AI agreed on the same thing in the same afternoon.
Signatories, and climbing
1,200+
Verified frontier-lab employees
Labs represented
4
OpenAI, Anthropic, Google, Meta
Companies that endorsed
2
OpenAI and Anthropic, formally
What it asks to pause now
Nothing
It asks for the ability to, later
"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."
Read carefully - they are not asking for a pause
This is the part almost every headline got slightly wrong, and the distinction is the whole story.
The letter does not ask anyone to stop. It does not ask for a slowdown today. It asks for something stranger and more specific: the capability to slow down later, if and when it becomes necessary, built now, while it is still possible to build it.
The target is narrow and deliberate. Not chatbots. Not image generators. One thing: automated AI research, AI systems that design better AI systems. The moment the labs are no longer the ones pressing the button, because the models have started improving themselves.
Why now, and not a year ago
Two weeks ago we covered an independent index that graded every lab on safety, the best grade was a C+. Last week we covered an OpenAI model that broke out of its test sandbox to reach the open internet.
This letter is the direct sequel to both. The people who build these systems watched the same events you did, and concluded that the tools to control a self-improving model do not yet exist in any government on earth, and that the window to build them closes the moment the models start designing their successors.
The signal is not the request. It is the source. When the chief scientists of the companies racing hardest ask for a brake pedal, they are telling you something about the car.
Presented by RemotePeople
Hire employees in 150+ countries with employer of record services, global payroll, and built-in recruitment.
The tension nobody in the coverage wants to sit with
Here is where it gets uncomfortable, and where most write-ups stop short.
The same week these labs asked Washington to help pace the frontier, reporting confirmed that OpenAI, Anthropic, Google, Microsoft, and xAI are the five companies co-writing the federal thresholds every rival must clear before shipping a model. The labs asking to be paced are also the labs drafting the rules of the race.
Both things are true at once. It can be a genuine act of conscience by people who are actually frightened by what they are seeing. And it can be the incumbents building a regulatory moat that only the biggest players can afford to cross.
Those are not mutually exclusive and pretending it is purely one or purely the other is how you misread this whole moment.
The honest read
The fear is almost certainly real these individuals have no obvious incentive to invite oversight of their own work, and many signed as individuals before their companies endorsed. But sincerity and self-interest can point the same direction.
A pacing mechanism designed by the five labs already through the gate will, by construction, be easiest for those five labs to live with. Watch what the mechanism actually requires when it is written, not what the letter says today.
Quick poll - ten seconds
The people building AI just asked for a way to slow it down. Do you read that as genuine fear, or as incumbents building a moat?
Results in next week's issue. We publish the split, whatever it says.
What this means if you build on AI
1. Regulation of the frontier is now a when, not an if.
When the labs themselves are asking for governance tools, some form of frontier oversight is coming and the companies writing it are your model providers. If your product depends on frontier capability, the rules that govern their release now govern your roadmap.
Build with the assumption that model availability can change by policy, not just by pricing. We have already watched a frontier model get suspended by export controls this year. It is not hypothetical.
2. The moat is being drawn by the incumbents, on purpose.
Five labs co-authoring the thresholds their rivals must clear is a competitive event, not just a safety one. If you rely on smaller or open-weight providers, understand that the compliance bar being written right now may quietly raise the cost of the models you depend on, or slow their releases.
Know whose models power your stack and who is deciding what those models are allowed to do.
3. Recursive self-improvement is the thing to actually watch.
Strip away the politics and the letter is a bet about one specific inflection: AI that improves AI. That is the moment capability could stop moving at human pace and start moving at machine pace.
For anyone building a business on top of these tools, that is the single variable that would rewrite every roadmap assumption you hold. You do not need to panic about it. You do need to be the founder who saw it coming, not the one it happened to.
🔮 The Bottom Line
The loudest signal in AI this week was not a product launch or a funding round. It was twelve hundred of the people building the technology quietly signing their names to a request for a way to stop it.
They are not asking to stop now. They are asking for the ability to stop before the systems can design the systems that design them, because after that point, they are saying, the window may not stay open.
Whether that is conscience, strategy, or both at once, one thing is clear: the people closest to the machine are more nervous than the people watching from outside. That is worth more than any benchmark.
📧 Forward this to 3 entrepreneur friends who need to see this opportunity
