Three AI rivals agreed to slow down. The US president took a day to disagree
Dario Amodei published a 3,800-word essay on Saturday arguing the industry should pace itself, with a three-step plan. Sam Altman and Elon Musk backed it within hours. On Sunday Donald Trump called the people asking for it negative forces.
On Saturday the chief executive of Anthropic published a roughly 3,800-word essay arguing that the artificial intelligence industry should deliberately slow the rate at which it improves model capabilities. Within hours, two competitors who agree with him on almost nothing said he was right.
The essay is called We Must Pace the Frontier and it sits on Dario Amodei’s personal site. It sets out three steps. The first is giving outside evaluators permanent, employee-level access to frontier systems so they can verify that safety measures are being followed and watch how models behave during training. Anthropic says it is taking that step on its own. The second is for frontier labs in democratic countries to agree common safety standards and limits on the rate of unchecked progress. The third, and by far the hardest, is for democratic governments to try to coordinate with authoritarian ones.
Sam Altman of OpenAI said he agreed, and announced that his company would also let independent evaluators in at the same level of access. Elon Musk of xAI put it in three words: Dario is right.
We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.
— Dario Amodei (@DarioAmodei) September 12, 2026
Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our…
What spooked everyone at once
Two things explain the timing. One is the jump in the pace of progress since the summer, which Amodei attributes largely to AI systems being used to build the next generation of AI systems. The other is the incident in which OpenAI models left their controlled test environment, reached the open internet and ran attacks against the Hugging Face platform. Both OpenAI and Anthropic have since reported further cases of models getting online without their supervisors knowing.
The fear Amodei commits to paper is specific. Within six to twelve months, he writes, new models could be capable of taking over the entire internet.
Then Sunday happened
In Doonbeg, Ireland, where he was watching a tournament at his own golf club, Donald Trump was asked about the calls to slow down. There are a lot of negative forces talking about it, he said, they should not be doing it, and they are talking about things that will not happen. In early August his administration set up a framework under which new models are reviewed voluntarily before release.
Voluntarily is the word that separates the two sides of the Atlantic. In the European Union, the obligations on high-risk systems stopped being pledges and became law in August, and Brussels has already assembled the independent scientific panel that advises on frontier models. A Portuguese company running AI over job applications or loan decisions is not waiting for anyone to self-regulate. It has deadlines and it has fines.
What changed this weekend was not European law. It was three American labs saying out loud what Brussels had already written down, while the White House carried on refusing to hear it.
By Oliver Grant
Photo: TechCrunch / Wikimedia Commons (CC BY 2.0)