Anthropic has disabled access to its two most capable artificial-intelligence models, Fable 5 and Mythos 5, after the US government issued an export-control order on Friday evening barring their use by any foreign national — including the company's own non-citizen employees. The directive, which the company says it received at 5:21pm on Friday, was implemented within hours, abruptly cutting off users outside the United States from the frontier systems.
The Trump administration invoked national-security grounds, treating the model weights themselves as a controlled export. Washington has long used export controls to restrict the sale of the advanced semiconductors that train and run large AI systems, but extending that authority to the models directly marks a significant escalation and, by Anthropic's account, the first time the tool has been turned on a US-developed model rather than the chips beneath it.
Anthropic said the order did not spell out the specific concern in detail. The company's understanding is that the government believes there is a method of bypassing — or "jailbreaking" — a safeguard that would otherwise prevent Fable 5 from being used to identify software vulnerabilities. Mythos 5, the full non-public model reserved for government agencies and select corporate partners, is unusually strong at surfacing flaws in code that have gone undetected for years, a dual-use capability prized for defensive hardening but dangerous if pointed at an adversary's systems.
Fable 5 is a newly released variant built on the Mythos technology, with its most sensitive cybersecurity and biotechnology capabilities deliberately blocked. The company said the safety measures in Fable 5 "have been extensively tested," and argued that competing systems, including OpenAI's GPT-5.5, possess broadly similar code-review abilities — making a model-specific ban on Anthropic an awkward fit for a capability that is not unique to it.
The practical effect was immediate and global. Developers, businesses and individual subscribers outside the United States lost access to the two models, and Anthropic staff who are not US citizens were locked out as well. The company has not said how long it expects the suspension to last or what would satisfy the government's concern.
The order lands amid escalating friction between Anthropic and the administration. The dispute traces in part to the company's refusal to allow the US military to use its models for fully autonomous weapons systems, a stance that put it at odds with parts of the defense establishment even as it continued to supply government customers.
Earlier this month Anthropic had also proposed a coordinated, industry-wide pause on the most advanced AI development, warning that without one humans risked losing control as capabilities accelerate. That public call for restraint sits uneasily alongside a government order that frames the same frontier capabilities as too dangerous to share abroad.
Export-control specialists note the move raises unresolved questions about scope and enforcement. Software does not move across borders like a crate of chips, and a directive that targets one company's models while leaving comparable systems untouched may prove difficult to sustain — both technically and as policy — if rivals continue to offer similar tools to overseas users.
For now, Anthropic says it is complying while it seeks clarity. The company stopped short of saying whether it would challenge the order, noting only that it disagreed with the restriction and that the letter it received did not explain the government's reasoning.