US Orders Anthropic to Pull Its Most Advanced AI Models

Anthropic has abruptly disabled worldwide access to its two newest and most capable artificial intelligence models, Fable 5 and Mythos 5, after the US government invoked national security export controls to bar all foreign nationals from using them. The company disclosed the order in a public statement on June 12, saying it received the directive that afternoon and had no practical choice but to shut the systems off for every customer to ensure compliance. Anthropic stressed that access to all of its other models, including its latest Claude Opus 4.8, is unaffected.

According to Anthropic, the directive suspends access to Fable 5 and Mythos 5 by any foreign national, whether located inside or outside the United States, a scope that extends even to the company's own non citizen employees. Because there is no reliable way to separate foreign nationals from US persons in real time across a user base numbering in the hundreds of millions, particularly on same day notice, the company concluded that a blanket global shutdown was the only way to comply. Anthropic said it received the letter at 5:21 p.m. Eastern Time and that the document did not spell out the specific national security concern behind it.

Multiple news organizations reported that the order originated with the Commerce Department, and Bloomberg, citing a US official who confirmed the letter, characterized the action by the Trump administration as an unprecedented move to keep frontier AI out of the hands of all foreign nationals. The timing is notable: the models had been made available to the public only days earlier, and Fortune reported that Anthropic recently filed confidentially for a public listing following a funding round that valued the company at 965 billion dollars, raising questions among observers about how repeated government restrictions might color investor sentiment.

Anthropic's account frames the episode as a disagreement over technical facts. The company said its understanding is that the government acted after learning of a method to bypass, or jailbreak, Fable 5's safeguards. Anthropic said it reviewed a demonstration of the technique and found that it surfaced only a small number of previously known, minor vulnerabilities, ones that other publicly available models can also discover without any bypass. To date, the company said, the government has provided only verbal evidence of a narrow, non universal jailbreak that essentially amounts to asking the model to read a specific codebase and fix software flaws, a capability Anthropic says is widely available from other systems, including the model it identified as OpenAI's GPT-5.5, and is used daily by the defenders who keep systems secure.

In its original launch materials, Anthropic had described a defense in depth strategy for Fable 5, acknowledging that perfect resistance to jailbreaks is likely not currently possible for any provider and that universal jailbreaks, ones that broadly unlock restricted capabilities, will probably be found eventually. The company said its aim was to keep jailbreaks either narrow or very expensive to produce, paired with monitoring to detect and shut down attacks quickly, and noted that this approach is why it required 30 day retention of customer data for the model. Anthropic said red teaming before launch, conducted with the US government, the UK AI Safety Institute, third party organizations, and internal teams over thousands of hours, found the model's safeguards substantially more effective than those of any previously deployed model, with no universal jailbreak discovered.

Anthropic said it is complying with the legal directive while openly disagreeing with it. The company argued that recalling a commercial model used by hundreds of millions of people over a narrow potential jailbreak sets a standard that, if applied across the industry, would essentially halt all new frontier model deployments. It pointed to its own previously stated policy position that the government should be able to block unsafe deployments, but only through a statutory process that is transparent, fair, clear, and grounded in technical facts, conditions it said this action did not meet. Anthropic apologized to customers, called the situation a misunderstanding, and said it was working to restore access as soon as possible, promising further detail within 24 hours.

The two models occupy the top of Anthropic's lineup. Fable 5 marked the first time the company released such an advanced offering to the public, made possible by new safeguards that block responses in specific high risk areas, while the Mythos line traces back to a Mythos Preview that drew attention from Wall Street and government officials in April for its cybersecurity capabilities and was limited to a small group of organizations under an initiative the company calls Project Glasswing.

Stripped of the specifics, this is a story about who holds the off switch for frontier AI, and the answer, at least for now, is the state. That a single export control letter could darken a flagship product for every user on the planet within hours illustrates how thoroughly national security law has come to sit atop the AI industry, and how a regime built to govern the movement of physical goods across borders is being stretched to cover software accessed through a browser. The foreign national framing is striking in its breadth: an export control that reaches a company's own employees inside the United States blurs the line between controlling what crosses a border and controlling who may use a domestic service.

It is worth noting the limits of what can be verified here. Almost the entire factual record so far comes from Anthropic itself, an interested party, while the government has said little publicly and, by Anthropic's account, did not detail its concern in writing. Independent confirmation of the order's existence has come from officials cited by outlets such as Bloomberg, but the substance of the security worry remains opaque, and readers should weigh the company's characterization of a mere misunderstanding accordingly. Looking ahead, the episode is likely to intensify a debate the industry has so far mostly deferred: what due process should govern a government's power to pull a deployed model, how providers should architect for the possibility of sudden shutdown, and whether the United States can restrict frontier systems without also constraining the open, global research culture that produced them. However this particular dispute resolves, the precedent it sets may outlast the models at its center.