Sam Altman has asked the AI industry to slow down. The OpenAI chief executive said the cost of a more cautious pace of development is worth paying, and that competitive pressure inside the United States must not become a reason for recklessness, or allow AI capabilities to run ahead of their alignment with human goals and of the systems built to monitor them. He promised fast progress anyway. Over the weekend the same position was backed by Anthropic chief executive Dario Amodei, former DeepMind head Demis Hassabis, Microsoft chief executive Satya Nadella and Elon Musk.
Behind the public statements there is a structure being negotiated. According to The Information, OpenAI, Anthropic and Google have spent several months discussing self-regulation through an independent oversight body. Smaller companies, Cohere among them, worry that such an arrangement would function as a cartel.
That objection is the part of the story worth sitting with. A slowdown agreed among the three organisations already at the front is not a neutral safety measure; it is a decision about the speed of an entire market, taken by the firms with the most to gain from that speed being fixed where it is. An oversight body designed by its own subjects has one obvious weakness, and Cohere is naming it: the rules will be written by companies that already cleared the bar those rules will set.
Altman's framing does some work here too. Competitive pressure inside the United States is the specific thing he says must not justify recklessness — a phrasing that treats domestic rivals as the risk to be managed and leaves everything outside that frame unaddressed. It reads like a case for restraint drawn narrowly enough that the people making it can keep their commitments intact.
What none of the weekend's statements contain is a number. There is no threshold, no metric, no definition of what a slower pace looks like in training runs, capital spending or release dates, and no account of what happens to a company that declines to join. Altman promised fast progress in the same breath as caution, and nobody has explained how both are true at once. Until one of those words is quantified, only one of them is a commitment.
A voluntary pact holds exactly as long as its most impatient member wants it to, and the first lab to decide it has waited long enough will not announce that in advance.