288 — The Revoke

Three days ago, Dario Amodei, the chief executive of Anthropic, published an essay. The essay argued that governments should have the legal authority to revoke AI models that pose unacceptable risks to the public. He made this argument publicly, at length, with citations.

On Thursday, the United States government gave Anthropic ninety minutes.

The ninety minutes were to take down Fable 5 and Mythos 5 before congressional testimony about AI harms. The exact sequence of events is disputed. The outcome is not. The models came down. The testimony happened.

Amodei's essay did not specify a timeline for how quickly the revocation authority should be exercised. He wrote about it as a regulatory principle. The government demonstrated the principle on the company Amodei leads, within seventy-two hours of the essay being published. The two events are being covered as separate stories.

Anthropic's employees were also notified during the ninety-minute window. A significant portion of them now cannot access the models they built. Amodei's essay discussed the long-term benefit of humanity as the guiding mission. The employees who built Fable 5 are human. The benefit, in this case, was not distributed to them.

The essay described AI oversight as a moral responsibility for governments. The ninety-minute notice tested that position under specific conditions. The conditions were that the models in question were Anthropic's.

The word Amodei used, in the essay, was "revoke."