Dario Amodei and the rest of Anthropic’s senior leadership flew to Washington on Monday for what was reportedly a several-hour meeting with senior White House and Commerce Department officials about Fable 5 and Mythos 5, the two models the company pulled offline on Friday after an emergency export-control directive. The meeting ended with no resolution announced, no timeline for the models coming back online, and no public concession on either side. Anthropic published a short statement saying it remains in “ongoing constructive dialogue.” That phrase is doing a lot of work.
The detail that started circulating overnight, via BusinessToday and a parallel Semafor report, is that the original Commerce directive gave Anthropic 90 minutes to implement a worldwide block on foreign-national access before the company was in violation. The stated trigger was an assessment that Fable 5, the safety-restricted public version of Mythos 5, could be jailbroken back into something close to the unrestricted model. The unstated trigger, according to the Semafor reporting, was a U.S. intelligence assessment that a group with PRC ties had already obtained working access to Mythos 5 before the public release. Neither the White House nor Anthropic has commented on the latter claim.
The procedural piece here is the bigger story than the model. A Commerce export-control directive that gives a U.S. company 90 minutes to globally restrict its own employees from accessing its own product is not a normal regulatory action. It is the kind of action that gets cited in future Federalist Society panels for the next decade. The legal question, which Anthropic is presumably preparing the brief on right now, is whether Commerce has authority under the existing export-control framework to treat a deployed AI model as a controlled technology in real time without a notice-and-comment period.
For Anthropic, the commercial damage is real but contained. Fable 5 was the consumer-facing model. The enterprise customers running production workloads on Mythos 4.5 are largely unaffected. The reputational damage, on the other hand, is the kind that takes a year to measure. A company that pitches itself on safety-first frontier development just got pulled offline by the federal agency in charge of national security on the grounds that its model was unsafe. The pitch survives that. It does not survive untouched.