Anthropic disabled Claude Fable 5 and Claude Mythos 5 for every customer worldwide on the evening of June 12, four days after their public launch, per the company's statement. A letter from Commerce Secretary Howard Lutnick, received at 5:21 PM ET that afternoon, subjected both models to export controls under national security authority, per Fortune. Anthropic complied. All other Claude models, including Opus 4.8, remained available.
The technical mechanism behind the order is worth getting right. A trusted partner of both Anthropic and the US government discovered a jailbreak that used Claude Fable 5, the consumer-tier conversational model, as the entry point to bypass guardrails between Fable and Mythos 5, Anthropic's cybersecurity-tuned reasoning model. The bypass unlocked Mythos's cyber capabilities through a Fable prompt, per MarkTechPost. The administration asked CEO Dario Amodei to fix the bypass or de-deploy. Amodei declined, arguing the jailbreak was narrow and would not defeat all of Fable 5's safeguards.
What the export-control order actually did is the second piece to get right. Commerce subjected the two models to export controls outside the United States and to foreign persons inside the country, per Al Jazeera. Because both models are delivered as hosted API services rather than downloadable weights, the practical compliance step was a global shutdown. There is no technical way to deliver an API-accessible model only to a citizenship-screened subset of paying customers worldwide. The shutdown was the only available compliance posture.
Bottom Line
This is the first government-forced disabling of a publicly deployed frontier model. The mechanism, export controls applied to a hosted service, is the precedent every other lab is now studying regardless of where they sit on the underlying disagreement.