A U.S. export control directive forced the AI company to disable Claude Mythos Preview and Fable 5 for foreign nationals worldwide, citing an unspecified national security concern.
On Friday June 12, 2026, Anthropic abruptly disabled two of its frontier AI models after receiving what the company called a "highly restrictive U.S. government order." That order is an export-control directive that bars foreign nationals from using the affected systems either inside or outside the United States. The company's public response, reported by Gizmodo on Saturday, framed the move as a misunderstanding it is working to reverse.
The action targets Claude Mythos Preview — internally designated Mythos 5 — a frontier model Anthropic itself rolled out in early April 2026 through a "consciousness-raising campaign" that framed the system as too dangerous to release publicly. A second model, Fable 5, was disabled in the same directive. Anthropic confirmed both names in a statement posted to its website on June 12, 2026.
Export controls are a category of trade restriction historically used on sensitive hardware, dual-use biotechnology, and encryption, and they normally operate through the Commerce Department's Bureau of Industry and Security. Gizmodo's report does not identify the issuing body or the specific rule invoked, and the underlying directive text was not available for independent review at the time of publication. What is on the record is Anthropic's own characterization: an export control citing an unspecified national security concern.
That characterization matters. The federal government is not on the record explaining which threat the directive is meant to address, and the "unspecified concern" language is Anthropic's paraphrase rather than an authoritative finding. Until the directive text or an official Federal Register entry becomes public, the actual legal mechanism, whether this is a new rule, a license requirement, a temporary denial order, or something else, remains an open question. So does whether other frontier-model providers are now operating under similar restrictions.
What is documented is the rhetorical ground the directive appears to stand on. Anthropic's April launch of Claude Mythos Preview was built around vivid catastrophic-risk language: a system card describing the model as relevant to catastrophic biological weapons development, deception, and containment-breaking behavior. The company did not release Mythos Preview to the public. It sampled the model, blogged about the worst-case scenarios it found, and built a separate program (Project Glasswing) explicitly to probe for those failure modes. By the company's own account, the system was being treated, in public, as an unusually dangerous artifact.
The export-control directive, as described by Anthropic, extends that treatment to the legal regime normally reserved for chips, biotech, and cryptography. If the directive's reach is what Anthropic says it is, a foreign national anywhere in the world is now prohibited from using Mythos 5 and Fable 5, even from a U.S. data center, even with a U.S.-based account, and even in jurisdictions with no domestic AI restriction of their own. That posture is closer to the way the U.S. treats advanced semiconductor designs than the way it treats consumer software.
Anthropic's own statement, posted to its website on June 12, 2026, says the company "believe[s] this is a misunderstanding and are working to restore access as soon as possible." That is a company claim about its own situation, not a government concession. The Bureau of Industry and Security, the Commerce Department, and the White House have not, on the record at the time of this article, confirmed or denied the directive's existence, scope, or rationale.
The Anthropic case is also the first time that lab-authored catastrophic-risk language has produced a concrete federal action that takes a frontier model offline. For roughly two years, frontier AI labs have been telling the public, and federal regulators, that their most capable models pose catastrophic and bioweapons-relevant risks. The next questions are empirical: which agency issued the directive, under what statutory authority, whether it singles out Mythos 5 and Fable 5 or applies to a class of frontier systems, and whether it survives the challenge now mounting from the company, the model's users, and the labs that will have to decide how literal their own safety language should be.