Commerce Department orders Anthropic to take Fable 5 and Mythos 5 offline worldwide
Amazon researchers had reported a technique bypassing Fable 5's safeguards; Anthropic disputed the order's rationale and said less capable models showed the same weakness.
- Security & misuse
- Government & policy
- Era-defining
The US Commerce Department issued an export-control directive ordering Anthropic to bar all foreign nationals — including non-citizen employees inside the United States — from accessing Claude Fable 5 and Claude Mythos 5, three days after their launch. The order arrived by letter at 5:21pm ET on a Friday with no prior notice. Because Anthropic said it had no reliable way to verify user nationality in real time, it disabled both models worldwide rather than attempt selective blocking.
The stated trigger was a jailbreak: officials said Amazon researchers had found a way to prompt Fable 5 into identifying and demonstrating software vulnerabilities, bypassing its safety classifiers, and that the same technique could unlock cybersecurity capabilities in Mythos 5 in limited scenarios. Anthropic disputed the government’s framing, calling the exploit narrow and noting its own testing showed less capable, already-deployed Claude models could be prompted into similar behaviour — meaning, it argued, that applying the same standard consistently “would essentially halt all new model deployments for all frontier model providers.” Only Fable 5 and Mythos 5 were affected; other Claude models remained available throughout.
The order landed amid an already tense relationship between Anthropic and the administration: Trump had ordered federal agencies to stop using Anthropic’s models in February after the company declined Pentagon contract terms permitting unrestricted use, including in autonomous weapons systems, and the Pentagon had separately designated Anthropic a “supply chain risk” in March, a designation Anthropic was contesting in court. Reporting attributed the initial complaint behind the June order to Amazon chief executive Andy Jassy, whose company is both an Anthropic investor and the discoverer of the jailbreak.
Access was restored globally on 1 July, after Anthropic deployed a revised safety classifier it said blocked the reported technique in over 99% of cases, with remaining flagged requests routed to the older Claude Opus 4.8. The 18-day shutdown was the first instance of a US export-control action being used to pull a domestically developed frontier model offline worldwide, rather than to restrict its export to specific countries.