Dario Amodei wants the industry to slow down, and several of his loudest rivals have now said they agree.
The Anthropic chief published an essay titled We Must Pace the Frontier, setting out three steps. First, Anthropic will give outside evaluators permanent, employee-level access to its systems, naming METR as an example. Reviewers would get desks, badges, laptops and permissions close to what internal risk teams hold, with a mandate to verify safety practices and report incidents inside training pipelines, not only on finished models.
Second, labs in democratic countries would agree on shared safety standards and limits on unchecked progress. Third, governments including China and Russia would join a global framework. Amodei called that last stage the hardest and said democracies must keep a technological lead while pursuing it.
Sam Altman endorsed the idea within hours. He said pacing had been a running topic inside OpenAI for weeks and that his company would grant independent evaluators similar access. Elon Musk and Demis Hassabis also backed parts of the plan.
Altman then told Fortune that OpenAI would not go public in 2026, calling it an ill-advised moment given the safety debate. The company had filed confidentially and once aimed at late 2026, but reporting has pointed to a shift toward 2027.
Critics remain unconvinced. Some researchers argue recursive self-improvement is a weak premise, and journalists warn that voluntary standards can amount to regulatory capture.
