For years, the people most worried about AI asked for caution, and some wanted an outright pause. Anthropic was one of the loudest voices urging the world to take the danger seriously, even as it kept building. It told the world its most advanced model was simply too powerful to release.
Last week, the pause arrived. Not from the labs, or the safety researchers who asked for it, but from the White House, on a Friday evening, as an instruction to shut access down almost overnight. Anthropic disagreed and said so. The company that argued loudest for the danger to be taken seriously did not get to choose what that would look like.
I have spent my career in change programmes, and this is one of the oldest lessons there is. When you spend years arguing that something is dangerous, you do not get to decide who acts on it, or how hard, or when. You hand someone else the hammer, and then you live with the swing.
There is no clean villain here. You can think the models are risky and still think the response was clumsy. Both can be true.
So ask carefully. Name what you want, not only what you are against. The world has a habit of granting the wish and not the intention.
Anthropic asked the world to take the danger seriously. On Friday 12 June, it did.

Source: Reed Albergotti, “US limits use of Anthropic AI models Fable 5 and Mythos”, Semafor, 12 June 2026. Anthropic published its response to the directive on anthropic.com.