The US government has ordered the artificial intelligence company Anthropic to bar all foreign nationals from its two most powerful AI models, forcing the company to shut them down entirely on Friday.
The Commerce Department issued the export control directive at 5:21 p.m. Eastern time, citing national security and telling Anthropic to cut off access to Fable 5 and Mythos 5 for any foreign national, whether inside or outside the United States, including the company’s own employees who are not US citizens. Anthropic said the only way to comply was to switch the models off for every customer. Its other systems, including the Claude Opus 4.8 model, stayed online.
Anthropic said the letter gave no detail on the concern. By its account, officials believe someone found a way to jailbreak Fable 5, the public version built on the more powerful Mythos system, to reach cybersecurity abilities the safeguards are meant to contain. The company called the order a “misunderstanding” and said it was working to restore access.
It also disputed the severity. Anthropic said the technique it was shown surfaced only minor, already known flaws, and that the jailbreak was narrow rather than universal. It argued that other public models, including OpenAI’s GPT-5.5, could be pushed to do much the same. The Commerce Department did not immediately comment.
Friday’s order deepens a fight that began early in 2026. In March, the Pentagon designated Anthropic a supply chain risk, a label usually reserved for contractors tied to foreign adversaries, and said the firm threatened national security. The dispute followed a breakdown in contract talks: Anthropic refused to let its Claude models power autonomous weapons or mass surveillance of Americans, while the Pentagon wanted access for all lawful purposes, saying a private company cannot dictate how the government uses technology in a security emergency.
Anthropic sued the Defense Department and other agencies, calling the campaign unlawful retaliation for its safety stance. A federal appeals court refused to pause the designation in April, though a separate judge had blocked a wider ban on government use of Claude.
The two models at the center of Friday’s order are recent arrivals. They built on Claude Mythos Preview, disclosed in April, which Anthropic kept out of general release because of its cybersecurity power and offered instead to a small group of firms through a program called Project Glasswing.
The episode shows how far governments will go to control frontier AI, especially systems with cybersecurity and defense uses, and how badly that sits with developers that set their own limits. Anthropic said its remaining models work as normal and that it hopes to bring the two back once the matter is settled.
