Anthropic’s two flagship A.I. models, Fable 5 and Mythos 5, entered their seventh day offline on Thursday, the longest forced shutdown of a frontier commercial A.I. product in the brief history of the industry and the first time the Commerce Department has invoked its 2018 Export Control Reform Act authority against a commercial A.I. model, according to an export-control expert cited by Reuters.

The directive arrived by letter at 5:21 p.m. Eastern on Friday, from Commerce Secretary Howard Lutnick to Anthropic chief executive Dario Amodei, threatening “prompt criminal and civil penalties” if exports continued. Anthropic complied within hours, then issued a statement calling the action a “misunderstanding” and noting that the capabilities at issue were “widely available from other models, including OpenAI’s GPT-5.5.” Senior Anthropic technical staff have met with Commerce officials nearly every day since, per a person close to the company quoted by Reuters.

What looks on the surface like an export-controls dispute is, on closer inspection, the culmination of a year-long deterioration in Anthropic’s standing in Washington. The Washington Post reported that President Trump ordered federal agencies in February to stop using the company’s models after Anthropic declined Pentagon contract terms permitting autonomous weapons use. By March, the Defense Department had designated the company a “supply chain risk.” The Friday letter is what that posture looks like once it acquires a legal instrument.

The proximate trigger appears to be Project Glasswing, Anthropic’s restricted-access program covering roughly 150 firms, which the company expanded to include a Korean telecommunications company “suspected of having ties to China,” according to White House officials cited by the Post. SK Telecom invested $100 million in Anthropic in 2023. One unnamed official put it bluntly: the company “expanded it too far and wide.”

The cybersecurity industry has tried to intervene. On Sunday, more than 80 executives signed an open letter to Lutnick and National Cyber Director Sean Cairncross asking the administration to lift the restrictions. Researcher Peter Girnus, writing on X, captured the structural irony: “If you describe your product as a munition in every press release, eventually a government takes you at your word.”

That’s the genuine novelty here. The safety-coded posture that Anthropic cultivated to differentiate itself from OpenAI, Nvidia, and Adobe in the policy conversation has now been ratified, in the most literal possible way, by the state. The company built the regulatory frame; the regulator picked it up.

Sources