Three days after Anthropic launched the most capable AI models it has ever made publicly available, the U.S. government ordered them taken down. On June 12th, the Commerce Department issued a
Three days after Anthropic launched the most capable AI models it has ever made publicly available, the U.S. government ordered them taken down.
On June 12th, the Commerce Department issued an export control directive instructing Anthropic to suspend all access to Claude Fable 5 and Claude Mythos 5 for any foreign national — whether located inside or outside the United States, including foreign national employees working at Anthropic itself. This appears to be the first government-forced takedown of a publicly deployed frontier AI model.
Anthropic received the directive at 5:21pm Eastern Time. The letter did not provide specific details of the national security concern. Faced with a directive it could not comply with selectively — there is no practical mechanism for filtering foreign nationals from domestic users in real time at the model access level — Anthropic took the only available path: shutting both models down entirely for every customer worldwide. Access to all other Anthropic models was not affected. Claude Opus 4.8 and the rest of the Claude family remained fully operational.
What Fable 5 and Mythos 5 Actually Were
The models at the center of this unprecedented government intervention were not routine product updates. Both models stemmed from Claude Mythos Preview, a highly advanced model intended for security research, which was capable of finding security bugs and flaws. Access to Mythos Preview was initially limited to a small group of companies and research partners through Project Glasswing. Mozilla alone said it resolved hundreds of vulnerabilities as a direct result of using Mythos Preview.
When Anthropic launched Fable 5 and Mythos 5 on June 9th, the architecture was split deliberately. Mythos 5 retained its full capabilities but remained restricted to Project Glasswing participants. Fable 5 was the version made publicly available — a Mythos-class model with guardrails and safety restrictions designed to make it suitable for general use. Anthropic described Fable 5 at launch as state-of-the-art on nearly all tested capability benchmarks, with exceptional performance across software engineering, knowledge work, vision, and scientific research. The longer and more complex the task, the larger Fable 5’s lead over previous models.
The Jailbreak Claim That Triggered the Order
The Commerce Secretary sent a letter to Anthropic CEO Dario Amodei instructing that the Mythos 5 and Fable 5 models would be subject to export controls to any location outside of the U.S. and to all foreign persons within the country. An administration official told Axios the Commerce Department decided to take the action after another company claimed it was able to jailbreak Mythos, alarming the administration about possible national security risks.
Anthropic’s own assessment of the jailbreak claim is considerably less alarmed. Anthropic said it believed the jailbreak the government was citing was a narrow one that would unlock Mythos’s cybersecurity capabilities in only one specific instance and not a universal one that would defeat all of Fable 5’s safeguards. It also said it believed the same jailbreak could be used to elicit similar capabilities from other publicly available models, including OpenAI’s GPT-5.5, that are not subject to similar national security export controls.
“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,” Anthropic wrote in its blog post.
“We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible,” Anthropic stated publicly on X.
Why the Scope Required a Full Shutdown
The mechanics of why Anthropic had to disable the models for everyone — not just foreign nationals — reveal a structural limitation that the directive’s authors may not have fully considered. Distinguishing between U.S. and non-U.S. users in real time at the model access layer is not a technical capability Anthropic has built into its infrastructure. An export control directive targeted at foreign nationals, applied to a model already deployed to hundreds of millions of users globally, leaves only one compliant option: taking it down for everyone.
This is not an edge case. It is the fundamental challenge that export controls face when applied to software products distributed over the internet rather than physical goods. The directive effectively forced Anthropic to choose between compliance and service continuity — and compliance won. Every customer worldwide, including domestic U.S. users with no connection to the national security concern, lost access to two of the most capable AI models ever deployed.
What This Means for Anthropic’s Future
Anthropic confidentially filed for a public listing earlier this month. A recent funding round valued the company at $965 billion. The government export control decision could make investors less enthusiastic about an Anthropic IPO, causing them to question whether it will be able to stay at the cutting-edge of AI model development if the government continues to single out its models for various restrictions.
The precedent set by this directive extends well beyond Anthropic. If the U.S. government can issue an export control order that forces the immediate global withdrawal of a deployed frontier AI model based on a third-party jailbreak claim — without providing specific details of the security concern or waiting for the developer to assess the risk independently — then every frontier AI lab operating in the United States now faces the same potential exposure. Building and deploying the most capable models in the world may come with the risk of having them withdrawn at short notice at the direction of national security authorities.
Anthropic has stated it is working to restore access as quickly as possible. The path to that restoration remains unclear.
