AINEWS
A US export-control order pulled Anthropic's newest AI models off the internet within 72 hours
Fable 5 and Mythos 5 went offline globally on June 12 after a directive barring 'foreign nationals,' including Anthropic's own non US staff.
Fable 5 and Mythos 5 went offline globally on June 12 after a directive barring 'foreign nationals,' including Anthropic's own non US staff.
On June 12, 2026, Anthropic received a US government export-control directive requiring the company to cut off all "foreign nationals" from Fable 5 and Mythos 5, its two newest AI models, according to BleepingComputer's reporting on the directive. Because the directive applies inside and outside US territory, and because "foreign national" includes Anthropic's own non-US employees, the company's only compliant option was to take both models offline for every user worldwide.
The timing is what makes the order legible. Fable 5 had been free to all Pro, Max, and Enterprise customers for three days. It launched on June 9, 2026, was scheduled to remain free through June 22, and had been promoted to "millions" of users per Anthropic's framing, the same BleepingComputer account reports. By June 13 it shows "currently unavailable" to anyone who tries to load it. All other Anthropic models, including Claude Opus 4.8, remain accessible.
The lever, not the model names, is the story.
US export controls are the legal regime the government uses to restrict who can use certain US-origin technology, traditionally applied to physical goods like advanced chips and to encryption software. Applying that regime to a deployed AI model that is already running on company servers is a new posture. The directive, as paraphrased through Anthropic, cites "national security" authorities, reaches users outside the United States as well as inside it, and captures the company's own foreign-national workforce in the same net. That is the structural break: a single order aimed at adversaries can be operationalized only by removing the service from allies and from the company's own staff.
The technical basis is contested and was apparently never shown in writing. The directive reportedly turned on a claim that Fable 5 could be "jailbroken," meaning prompted in a way that bypasses its safety filters, to produce information useful to foreign threat actors. Anthropic's public position, as paraphrased by BleepingComputer, is that the alleged jailbreak consists of asking the model to read a specific codebase and identify or fix software flaws, a behavior the company says other publicly available models can also produce. The government has not, on the public record, released the written evidence that would let outside reviewers weigh that rebuttal.
The asymmetry is concrete and worth naming. Anthropic's own non-US engineers and researchers, who helped train and deploy Fable 5, lost access to it on the same directive that targeted foreign adversaries. Users in allied countries who had been using the model during its free-tier window lost it too. The order was not selective: there was no compliant way to fence off "foreign nationals" inside a model that anyone in the world could query, and the global takedown is the only operationally honest response.
What to watch next: whether the US government publishes the technical basis for the directive; whether Anthropic receives a narrower order that lets it serve non-US users under some access control; and whether other frontier-model providers receive similar requests, and disclose them the same way.