The U.S. Commerce Department ordered Anthropic on June 12 to suspend access to its Fable 5 and Mythos 5 models for all foreign nationals, anywhere in the world. 

Unable to filter users by nationality in real time, Anthropic pulled both models for every customer globally. The trigger was a reported jailbreak capable of unlocking Mythos’s offensive cybersecurity capabilities.

That action is the near-term story. The bigger one is what Anthropic CEO Dario Amodei said at a financial services event in May: there is a six-to-twelve-month window before Chinese AI reaches the same capability. Once Chinese labs release comparable models as open source, export controls become irrelevant — weights are public, safety guardrails become optional, and no regulatory action claws them back.

Mythos Preview, announced April 8, autonomously wrote working exploits for thousands of zero-day vulnerabilities — including a 17-year-old remote code execution flaw in FreeBSD’s NFS server and a 27-year-old crash bug in OpenBSD. Engineers with no security background could prompt it overnight and wake to a functional exploit. Anthropic restricted access to roughly 40 trusted institutions through a program called Project Glasswing and committed $100 million in usage credits to coordinated defensive work.

Two days before the ban, Amodei published “Policy on the AI Exponential,” demanding binding federal regulation of frontier AI and government authority to block unsafe model releases. Washington obliged — starting with his own company. Anthropic disputes the rationale: the jailbreak was narrow, it argues, affects one specific use case, and the same technique works on OpenAI’s GPT-5.5 without triggering equivalent controls. “If this standard were applied across the industry, it would essentially halt all new model deployments for all frontier providers,” the company wrote. Negotiations with U.S. officials began June 15 with no restoration date set.