Washington Orders Anthropic to Cut Off Foreign Access to Its Top Models

The United States government, citing national security authorities, issued an export-control directive on June 12 ordering Anthropic to suspend all access to its Fable 5 and Mythos 5 models by any foreign national, whether inside or outside the country, including Anthropic's own foreign-national employees. The company could not restrict access by nationality at the level of individual requests quickly enough, so it disabled both models for every customer in order to comply.

This is a significant escalation in how the state treats frontier systems. Until now, export controls targeted the inputs to artificial intelligence, the advanced accelerators and fabrication tools, on the theory that restricting computing power slows adversaries. Restricting a model that is already deployed and serving traffic treats the trained weights and the inference endpoint themselves as the controlled item, a change of category that Al Jazeera reports has no clear precedent.

The stated trigger is narrow and, by Anthropic's account, supported by little evidence. The company says the government provided only a verbal description of a potential jailbreak. That bypass reportedly amounts to asking the model to read a specific codebase and fix software flaws in it, which would make Fable 5 useful for discovering or exploiting vulnerabilities. Anthropic notes that the directive arrived without written specifics and that it describes the bypass as non-universal. Fortune reports that the remaining Claude models are unaffected.

Consider who benefits if the claim holds. If a frontier model can reliably find exploitable flaws in arbitrary code, the case for restricting it resembles the case for controlling other dual-use capabilities, and the government has an interest in acting quickly. If the underlying jailbreak is as narrow as Anthropic suggests, a blanket suspension of two flagship models for all foreign users is a large cost imposed on the basis of unverified, verbally communicated evidence. What is verified is the directive and the shutdown. What remains asserted is the severity and breadth of the capability that justified it.