On June 12, 2026, the US government ordered Anthropic to shut off access to two of its newest AI models, Fable 5 and Mythos 5, for every customer around the world. The order came through an export control directive that cites national security authorities.
The directive bans access to the two models by any foreign national, whether inside or outside the United States. That includes Anthropic's own foreign national employees. To stay compliant, the company said it had to disable both models for all users, not just foreign ones.
Anthropic stressed that access to its other models is not affected. Only Fable 5 and Mythos 5 are being pulled.
What the Government Says Is Wrong
According to Anthropic, the letter arrived at 5:21pm Eastern Time and did not spell out the specific national security concern. The company said its understanding is that the government believes someone found a way to bypass, or "jailbreak," Fable 5.
Anthropic said it reviewed a demonstration of the technique. It described the method as asking the model to read a specific codebase and fix any software flaws. The company said the vulnerabilities found were minor, previously known, and simple enough that other publicly available models can find them too.
The company also said it validated a report it believes is the basis of the directive. It claims the level of capability shown is widely available from other models, and pointed to OpenAI's GPT-5.5 as a reported example. Anthropic said this kind of capability is used every day by the security defenders who keep systems safe.
The Safeguards Behind Fable
Anthropic laid out its position on Fable's safety measures, which it first described when the model launched. The company said it built strong safeguards aimed at reducing misuse for cybersecurity tasks, and that some users had complained the limits were too broad.
Before launch, Anthropic said it worked with the US government, the UK AI Safety Institute, outside organizations, and internal teams to test Fable's defenses for thousands of hours. It said no tester has yet found a "universal jailbreak," meaning a method that broadly unlocks the model's blocked capabilities.
The company admitted that perfect resistance to jailbreaks is likely not possible for any AI provider today. It said every safeguard in the industry can be beaten in narrow, specific cases, and that broader bypasses will probably be found at some point. To manage that, Anthropic said it uses a "defense in depth" approach, including a 30-day data retention policy that lets it research and shut down attacks.
Why Anthropic Is Pushing Back
Anthropic said it is complying with the legal order and removing access. At the same time, it openly disagreed with the decision. The company argued that a narrow, non-universal jailbreak should not be reason to recall a commercial model used by hundreds of millions of people.
The company warned that if the same standard were applied across the industry, it would essentially halt all new model launches by every frontier AI provider. Anthropic said it supports the government having the power to block unsafe deployments, but only through a process that is transparent, fair, clear, and grounded in technical facts. It said this action did not meet those standards.
Anthropic apologized to its customers for the disruption, called the situation a misunderstanding, and said it is working to restore access as soon as possible. The company said it would share more details within 24 hours.
