Trump rejects the AI slowdown sought by Amodei, Altman and Musk: 'whoever wins AI, wins'

By Pasquale Pillitteri · September 13, 2026.

Donald Trump on September 13, at a press conference in Dublin, rejected the joint appeal from three of the men leading the AI race to slow it down. "Whoever wins AI wins," he said, and dismissed the safety warnings as noise from "negative forces" that bring up "things that are not going to happen". The real reason he gave was geopolitical: the United States is ahead of China and has no intention of surrendering that lead.

The striking thing is not Trump's response, consistent with an administration that has spent months treating AI as a national security asset. The striking thing is what prompted it. On September 12, Dario Amodei, Anthropic's chief executive, published the essay "We Must Pace the Frontier", in which he asks labs to lengthen the time between one capability leap and the next to leave room for safety testing. It is not a moratorium: it is about metering the pace, not halting it. Sam Altman (OpenAI) replied that he agrees with Amodei, and Elon Musk (xAI) settled it in three words: "Dario is right". Three companies that normally fight over researchers, chips and contracts, aligned for once on the same point.

In between, Microsoft charted a third way. Satya Nadella wrote that any pursuit of superintelligence "has to rest on the principle that, if the AI we build does not help humanity and is not under human control, it is not worth pursuing", but in the same sentence he called for accelerating and spreading the benefits. It is the idea of "humanist superintelligence" that Microsoft AI chief Mustafa Suleyman has been championing: systems as tools shaped by human intent, not as a race to stop or slow down, but to redirect. It is worth recalling this with the same caution we applied to Mark Zuckerberg's "superintelligence for everyone" manifesto: when a tech giant appropriates the vocabulary of safety, real commitment must be separated from brand positioning.

Our reading is that this episode matters less for what each side says than for what it reveals about who is really in charge. Amodei, Altman and Musk have written essays and posted messages; Trump settled the debate in a single sentence, from an official tour, without needing to argue it any further. The power to set the pace of AI does not lie with the labs that build it, but with whoever controls public contracts, export licenses and the national security framework. Today that is Washington, not San Francisco.

The two opposing positions also have a weak point worth flagging without hedging. Trump's is treating safety as a peacetime luxury: if even part of the three chief executives' fears turned out to be well founded, competing with China would not be a reason to accelerate blindly, but exactly the opposite, a reason not to get it wrong. Amodei's is one of feasibility: a voluntary slowdown only works if every lab abides by it at once, and the history of industrial pacts teaches that it takes only one cheater for the rest to have incentives to follow. Without an external referee, national or international, "pace the frontier" risks remaining a declaration of intent among competitors.

It is also worth questioning the geopolitical premise on which Trump builds his entire argument. As sector context, the data we track in our comparisons show that the distance between frontier American models and Chinese open-source ones —GLM, Qwen, DeepSeek, Kimi— has narrowed rapidly over the past year. "We are ahead of China" is itself a claim that would have to be verified against hard benchmarks before being used to justify decisions, not a certainty to be taken for granted. That the White House's decisive argument rests on an increasingly contested lead should worry those calling for a slowdown as much as those calling for acceleration.

The other end of the regulatory pendulum has also swung this week: Senator Bernie Sanders has introduced a bill that would impose up to 20 years in prison for building or deploying superintelligence without authorization. It is proof that, when political power decides to intervene, it can do so with the same abruptness with which Trump waved the problem away, only in the opposite direction. Neither reaction, the flat rejection or the maximum criminal threat, is a substitute for governance that distinguishes demonstrated capability from hypothetical risk.

In the short term, what is clear is that AI's speed is no longer decided solely by the labs competing for talent and compute, but by the governments that hand out contracts, impose export controls and decide what counts as a threat. It is a concentration of decision-making power that should be watched as seriously as we watch technical risks. In the long term, however, this kind of friction, between those who want to go slower so as not to break anything and those who want to go faster so as not to lose the race, is precisely the mechanism that, well managed, can prevent the derailments that would ruin the underlying goal: an AI that cures disease, extends healthy life and distributes abundance instead of imposing it out of fear or haste.