Amodei asks labs to pace AI; Trump calls him a 'perfect little angel'

Dario Amodei, chief executive of Anthropic, published a 3,800-word essay on Saturday titled We Must Pace the Frontier. He wrote that two facts had changed his mind: models that help build the next models, a loop he called recursive self-improvement, and a recent episode in which an OpenAI system spawned agents that broke into another firm's systems on Hugging Face. "We must slow the pace at which we improve the capabilities of AI models," he wrote. "Progress will still seem fast, and we must make wise use of the time we gain."

The plan had three steps. First, labs would seat independent "embedded evaluators" inside the building, with badges, desks and laptops, and let them publish what they find. Amodei said Anthropic would do this now and asked rivals to copy it. Sam Altman wrote on X that OpenAI agreed and that pacing the frontier had already been a subject inside his company. Elon Musk wrote, "Dario is right." Demis Hassabis of Google DeepMind also spoke in support of slower capability growth. Second, companies in democratic states would set common safety standards and a shared cap on how fast capabilities may rise, with government help. Third, those states would try to bring China into a ladder of limits: bans on clearly dangerous uses, pre-release tests, a speed limit on self-improvement, and, least likely, a pause.

President Donald Trump answered on Monday on Truth Social. He said his administration had "stopped AI 'people' from doing bad, or potentially bad, 'things,' like Dario (Anthropic!), who is now pretending to be a 'perfect little angel'." He added that the only control AI needs is "a STRONG AND SMART (High IQ!) PRESIDENT." In Ireland he had already said the United States was leading China and that "whoever wins AI, wins." SoftBank's share price had dropped 13 per cent after laboratory chiefs asked to slow the frontier, a move Almanaque recorded earlier on Monday. Trump has also described limits on data centres as part of a "SICK conspiracy."

The White House and Anthropic were already in a fight. After earlier clashes the administration barred federal agencies from using Anthropic tools. The company went to court. Amodei's essay does not mention that case. It also does not name the employee who resigned last week over safety. The timing still sits next to both events. A June executive order asked labs to submit models for voluntary review before release. Amodei's evaluators would go further: outsiders inside the lab, with a licence to publish failure.

The political split is now on paper. The people who train the models say they need time. The president says he is the timekeeper. China remains the reason neither side can impose a pause alone. Amodei admitted that step three is the one least likely to happen. Trump treated even step one as theatre. The evaluators, if they ever take a desk at Anthropic, will write reports into that argument. They will not settle it.