Inside the U.S. Order That Forced Anthropic to Kill Its New AI Models for the Whole World

"The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance," the company stated on its official blog and on X . Access to its other AI models, including the widely used Claude chatbot, remains unaffected .

What Sparked the Government's Action

The government's national security concerns center on what officials described as jailbreak vulnerabilities and the models' potential to conduct autonomous cybersecurity research . According to reports, the government provided only verbal evidence of a potential narrow, non-universal jailbreak—essentially a method of asking the model to read a specific codebase and fix software flaws .

Anthropic's own earlier safety warnings about its models may have inadvertently contributed to the government's alarm. The company had previously acknowledged that Mythos 5, its most capable model intended for restricted use, presented significant cybersecurity risks, leading the White House to set up protections for federal agencies that were given special access . Fable 5 was released publicly just days before the ban as a safer version trained to refuse harmful requests, but the government's swift move suggests it was not convinced the safeguards were sufficient .

This is the first time the U.S. government has classified a specific large language model as a national security concern and used deemed-export rules to restrict access even to foreign nationals working on U.S. soil . Commerce Secretary Howard Lutnick personally communicated the export controls to Anthropic CEO Dario Amodei via letter .

The Unprecedented Inclusion of Anthropic's Own Employees

The directive's reach into Anthropic's workforce is what made compliance so absolute. Government officials explicitly stated the ban applies to "any foreign national, whether inside or outside the United States, including foreign national Anthropic employees" .

This meant that even researchers and engineers working on the Glass Wing project—Anthropic's internal initiative for frontier safety—who are not U.S. citizens or permanent residents were forbidden from accessing the models . With the restriction applying inside the company's own walls, Anthropic could not simply limit external customer access and continue development internally. The only path to compliance was a complete global shutdown until technical or legal solutions could be worked out .

The episode marks a dramatic application of U.S. export control law to artificial intelligence, treating advanced models more like weapons or sensitive defense technology. The deemed-export rules mean that sharing controlled technology with a foreign national inside the U.S. is legally equivalent to exporting it abroad .

Anthropic's Response: This Is a "Misunderstanding"

Anthropic swiftly complied but made clear it believes the government got this wrong. "We believe this is a misunderstanding and are working to restore access as soon as possible," the company said in its public statement .

The company has emphasized that the jailbreak evidence it received was narrow and not universal, suggesting the government may be overestimating the practical risks . In the meantime, the company has promised to keep all other Claude models online while it negotiates a path to restoring Fable 5 and Mythos 5 .

As of now, the unprecedented order stands, and the two most powerful AI models ever to be pulled from public access for national security reasons remain dark. How long that lasts—and what precedent it sets for other frontier AI companies—will be the next chapter in the rapidly intensifying debate over who gets to use the most capable AI systems.