The durable event is not that Claude Fable 5 and Mythos 5 went dark; it is that a live frontier model was pulled into export-control logic and the control variable was the user’s status, not the user’s location. Anthropic’s public notice says the US directive suspended access for “any foreign national, whether inside or outside the United States, including foreign national Anthropic employees,” and that the “net effect” was that Anthropic “must abruptly disable Fable 5 and Mythos 5 for all our customers.” [E1]
That makes the case legally stranger and more important than a normal country-block. Reuters reported that Commerce used authority under the 2018 Export Control Reform Act “for the first time,” according to an export-control expert, to block foreign access to Anthropic’s most advanced models; the same report said US controls had long focused on “the chips and tools that power AI rather than on restricting foreign access to AI itself.” [E2]
The national-security theory is not cosmetic. Reuters, single-sourcing the non-public Commerce letter, reported that Secretary Howard Lutnick ordered suspension “to destinations worldwide and all foreign nationals, wherever located,” because Commerce feared the models could be deployed by military-intelligence users in China, Russia or other countries of concern. [E3]
The legal hinge is familiar even if the product is new. Under the EAR’s deemed-export rule, releasing controlled technology or source code to a foreign person in the United States can be treated as an export to that person’s country of citizenship or permanent residency, and “foreign person” is synonymous with “foreign national.” [E4]
Commerce also has a military-intelligence hook. EAR §744.22 permits license requirements when there is an “unacceptable risk of use in, or diversion to” a military-intelligence end use or end user, including users “wherever located.” [E5]
The problem is that cloud AI systems were not built like chip shipments. A wafer can be stopped at a port, and cryptographic technology can be controlled through licensing and release rules; a frontier-model API must decide, at login speed and at enterprise scale, whether each human, contractor, employee, customer account, downstream application and possible beneficiary is a US person. Anthropic’s own notice implies that this citizenship-level filter was not operationally separable from ordinary service delivery, so a targeted control became a global kill switch. [E1]
The hard case for the order is that frontier models are not just speech tools or consumer software when they can be used as operational accelerants by hostile intelligence services. The hard case against it is that a sudden US-person gate on a deployed model teaches foreign developers, enterprises and governments that reliance on US frontier APIs can disappear through a letter, and that demand will migrate toward non-US substitutes such as Qwen, Kimi or DeepSeek when reliability becomes a sovereignty question rather than only a model-quality question. [E2]
The immediate dispute has not been resolved. Reuters reported that US officials met Anthropic on Monday 15 June to resolve the export-curb dispute, but as of 16 June access had not been restored; Reuters also reported that the European Commission remained in contact with Anthropic after the disablement, confirming that the blast radius had become transatlantic rather than merely company-internal. [E6] [E7]