The U.S. Commerce Department sent Anthropic a letter Friday invoking an obscure export control rule that barred non‑American users, including the company’s own staff, from accessing its flagship models Fable 5 and Mythos 5. Citing an unnamed national‑security concern, the directive forced the startup to shut both models down for all customers to stay compliant. The firm received no public copy, so details remain vague.

Security researcher Katie Moussouris obtained a private draft from Anthropic describing an Amazon‑led study that claimed a guard‑rail bypass in Fable 5. She noted the exploit required asking the model to “fix” code rather than merely review it, and argued that such behavior should not trigger export controls. Dozens of experts have since called the order hasty and dangerous.

Anthropic’s forced shutdown demonstrates that U.S. regulators can halt advanced AI services without judicial review, setting a precedent that could reach any domestic software firm. Industry observers warn that the move erodes confidence in American AI reliability for critical tasks. The administration has yet to clarify its rationale, leaving the tech sector to grapple with an increasingly uncertain compliance landscape.