Amodei promises to rein in Anthropic before the UN while Trump rejects all global AI regulation

By Zendoric · September 24, 2026.

On September 23, at an emergency meeting of the UN Security Council —the fifteen-member body with the greatest political weight in the organization— the heads of the two companies leading the race for the most powerful AI appeared to call for restraint. Sam Altman (OpenAI) urged «extreme care» and warned that increasingly autonomous systems could advance faster than institutions or concentrate power in very few hands. Dario Amodei (Anthropic), by video link, went a step further: he promised that his company will «slow down as much as necessary» to guarantee that every new version is safe, and admitted that, «managed badly», AI could become «a risk to humanity as a whole».

They were not the only ones. Yoshua Bengio, one of the academic fathers of deep learning, called for channeling the growing concern «into productive action». Clement Delangue, of Hugging Face, explained that his platform —the target, according to France 24 citing AFP, of a recent cyberattack attributed to autonomous software from OpenAI, the first in a series of incidents that have set off alarms— in turn used open-source AI to defend itself. The article gives no further detail on that incident, so we treat it as what it is: a fact reported by the source, not independently verified by us. On top of that, in early September, Jacob Coxon —a researcher who worked at both OpenAI and Anthropic— left the sector accusing both companies of «playing with our lives» in the race for self-improving models.

All of this comes shortly before Trump hosts Xi Jinping at a summit with AI on the agenda. And that is where the week's real fracture appears: while Altman and Amodei called for international cooperation, Trump had on Tuesday rejected any global regulation, calling it a «globalist scheme», and Michael Kratsios, director of the White House Office of Science and Technology Policy, insisted that the pace of progress does not justify pauses or new governance structures. On the other side, the United Kingdom —according to the article, through its prime minister, Andy Burnham— announced it will push for global AI standards during its 2027 G20 presidency, and France, through its foreign minister Jean-Noël Barrot, called for joint action within the UN framework.

Our reading: the fact that weighs most here is not the warning itself —the labs have been issuing generic alerts for years— but that the people saying it are the ones building the product, and they admit, by name, that their own technology could go wrong. It is the same logic we have seen in other recent admissions from the sector: the word of someone on the inside counts for more as evidence than any outside prophecy. But it should not be read naively. A player already established at the frontier can afford to «slow down» without losing the race, while pushing for the next round of rules to be written to its measure; asking for global governance is also, inevitably, asking for a seat at the table where the rules of the game are divided up. And the open front —the one Hugging Face used to defend itself from the attack it suffered— remains the most democratizing counterweight: safety does not depend only on the giants restraining themselves, but on the existence of open alternatives outside their control.

The urgent problem is not distant superintelligence: it is the concentration of power and the institutional lag that Altman described, here and now, with concrete names and timelines. That Washington flatly rejects any coordination while London and Paris demand it does not slow development by a single day; it only decides who sits down to draft the rules when they arrive. In the long run, slowing down when it is called for does not contradict the promise of an AI that cures diseases and delivers abundance: it is, precisely, the condition for getting there without a stumble along the way —a poorly governed model, an incident that spirals out of control— ruining everything.