Anthropic has abruptly disabled access to its flagship Fable 5 and Mythos 5 artificial intelligence models after the U.S. government ordered the startup to block foreign nationals from using the systems, citing national security authorities.
The startup said it received the export-control order at 5:21 p.m. ET on Friday, and it covers “any foreign national, whether inside or outside the United States,” including foreign-national employees. Anthropic responded by suspending access for all users, saying it was necessary to ensure compliance.
Other models remain unaffected, the company said in a statement.
Government Flags Potential AI Safeguard Bypass
The company said authorities did not provide detailed justification for the directive. However, Anthropic said it understands the government’s concerns relate to a potential method of “jailbreaking” Fable 5 — a technique used to bypass a model’s built-in safeguards.
Anthropic pushed back against the decision, saying the reported technique revealed only “a small number of previously known, minor vulnerabilities” and did not demonstrate a broad failure of the model’s safety protections.
“We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people,” the company said.
The directive lands amid an already fraught relationship between Anthropic and Washington. Earlier this year, the Pentagon designated the AI startup a “supply-chain risk” after it refused to relax restrictions on the military use of its models for autonomous weapons and domestic surveillance. Anthropic sued, alleging unlawful retaliation and violations of its free-speech rights, while multiple legal challenges remain ongoing.
Why Fable 5 Matters
The order also comes less than a week after Anthropic unveiled Fable 5 and Mythos 5, describing them as its most advanced AI systems to date.
Fable 5 was notable because it brought capabilities derived from Anthropic’s highly scrutinized Mythos program to a wider audience. Before launch, the company said the models underwent thousands of hours of testing involving government agencies, third-party researchers and internal teams.
Anthropic maintained that no researcher has identified a “universal jailbreak” capable of broadly bypassing the models’ safeguards and argued that the cybersecurity capabilities cited by officials are already available in competing frontier AI systems.
The company said it is complying with the directive while working to restore access, calling the action a “misunderstanding” that is not “transparent, fair, clear, and grounded in technical facts.”
Image via Shutterstock
