The US Commerce Department last week ordered Anthropic to bar foreign nationals from accessing its two newest AI models, Fable 5 and Mythos 5, citing an ‘unacceptable risk’ of diversion to military-intelligence end uses. A letter from Commerce Secretary Howard Lutnick to CEO Dario Amodei, obtained by US media, directed the company to obtain licences for any deemed export or reexport of the models. Unable to immediately filter users by nationality, Anthropic took the extraordinary step of shutting down both systems worldwide, leaving customers from Tokyo to Frankfurt locked out.

This action, taken on 13 June 2026, marks a historic departure. For decades, US export controls governed physical goods—chips, lithography machines, missile components. Now, a sufficiently capable AI model is treated as the legal equivalent of a crate of guided-missile parts. The order applies to any foreign national, whether based in Silicon Valley or Singapore, including Anthropic’s own non-US employees. Viewed from Washington, the move reflects a growing conviction that frontier AI models constitute dual-use technology too dangerous to circulate without strict oversight.

The ban lands at a particularly delicate moment for Anthropic, which has confidentially filed for an initial public offering that could value the company at nearly $1 trillion. The firm now faces what Fortune described as an existential crisis: the US government has effectively ‘blacklisted’ its flagship products twice, forcing investors to price in the risk that Washington can shutter revenue streams overnight. Yet the White House has also signalled flexibility, with a senior official indicating openness to direct talks with Amodei, suggesting the ban may serve as a negotiating lever rather than a permanent barrier.

In Paris, the French startup Mistral has emerged as a clear beneficiary. Its strategy of offering open-weight models that customers can deploy and control on their own infrastructure suddenly looks prescient, insulating it from the extraterritorial reach of US export law. From Mexico City, commentators highlight the irony: a measure intended to prevent foreign actors from exploiting cybersecurity flaws has instead forced a global blackout, disrupting legitimate commercial and research use. Analysts in London note that the episode accelerates the fragmentation of the AI market along geopolitical lines, with non-US firms likely to tout sovereignty and independence as competitive advantages.

The standoff between Anthropic and the Trump administration is more than a regulatory spat; it is a stress test for the emerging architecture of AI governance. If a negotiated settlement produces a licensing framework, it could become a template for managing other frontier models. If talks fail, the US may entrench a regime that treats advanced software as a strategic asset subject to the same restrictions as nuclear technology. Either way, the era of freely accessible, state-of-the-art AI has been abruptly curtailed, and the reverberations will be felt from boardrooms in New York to policy circles in Brussels and beyond.