Unpublished Letter Suspended Global AI Access for 3 Weeks: The Legal Risks of Export Controls

In a stark demonstration of U.S. regulatory might over artificial intelligence, the Commerce Department on June 12, 2026 ordered Anthropic to suspend global access to its newest frontier models, Fable 5 and Mythos 5, just days after their release. The order did not appear in the Federal Register, was not a published rule, and was delivered as an unpublished letter invoking the Export Control Reform Act and the Export Administration Regulations' military-intelligence end-use controls. For three weeks, the most capable AI models ever released were simply switched off for most of the planet, affecting any 'foreign person' worldwide. A partial carve-out for 'certain trusted partners' came in a follow-up letter on June 26, and Fable 5 access was restored with additional safeguards on July 1, but Mythos 5 remains limited to approved U.S. institutions. This episode follows a warning by French President Macron at the G7 summit in June, who, over a lunch with President Trump and the CEOs of Anthropic and OpenAI, cautioned that the unilateral ability to 'turn off the switch' would damage not only the economies relying on American AI but also the American AI companies themselves. The swift and opaque action has profound implications for the global AI landscape. While U.S. regulators can silently cut off access to private-sector models, Chinese labs continue to ship increasingly competitive alternatives. On July 16, Moonshot AI released Kimi K3, which became the first Chinese model to top a major coding leaderboard, surpassing both Fable 5 and OpenAI's GPT-5.6. Moonshot has promised to open-source the full model weights by July 27, making it free to download and run without any reliance on a U.S.-controlled cloud. Earlier, in mid-June, Z.ai had released GLM-5.2 under an MIT open-source license, and NIST's own evaluators assessed it as 'probably the most capable open-weight AI model' at its release. These Chinese models, though possibly still trailing in some areas, compensate by being cheaper, downloadable, and free of an American off switch. However, they are not without their own export-control risks, as Chinese regulations also govern cross-border transfer of advanced technology, and the provenance of open-source components can raise security and compliance concerns. For enterprises and governments worldwide, the choice is no longer simply about model capability. It has become a calculation of regulatory risk: reliance on U.S. models means exposure to sudden, non-transparent restrictions dictated by national security policy, potentially without any due process or public notice; turning to Chinese open-weight models may sidestep that singular dependency but introduces different legal and cybersecurity vectors. The three-week outage of Anthropic's models stands as a watershed, signaling that AI procurement must now integrate export control risk as a first-order consideration. As the U.S. and China vie for AI supremacy, the weaponization of access controls accelerates the fragmentation of the global AI market, pushing companies to dual-source or to favor models with more predictable availability—a trend that could inadvertently strengthen the open-source AI ecosystems of competitors.