The letter arrived at 5:21 p.m. Eastern Time on Friday, June 12. It was from the U.S. Commerce Department, and it ordered Anthropic to stop allowing any foreign national to access its two most advanced models, Fable 5 and Mythos 5. The directive covered foreigners outside the United States and inside it, including Anthropic’s own employees who are not U.S. citizens.
Anthropic had no mechanism to verify the nationality of every user in real time, so it did the only thing that guaranteed compliance: it disabled both models for everyone. Within hours, hundreds of millions of users lost access to the company’s flagship products, a shutdown the company itself described in a blog post as abrupt and disproportionate.
The episode is a case study in what happens when an AI company asks Washington for regulation and gets it, in a form it did not expect. Anthropic had spent years arguing for structured government oversight of frontier models. Its chief executive, Dario Amodei, had testified to Congress in favor of licensing regimes and had said publicly that the government should have the power to block unsafe deployments. The government, it turned out, was listening.
The trigger was a jailbreak, or what the government believed to be one. Officials told Anthropic they had learned of a technique that could bypass Fable 5’s safety guardrails, which were designed to prevent users from unlocking the powerful cybersecurity capabilities of the underlying Mythos model. Media reports said Amazon researchers had identified the bypass during testing and raised concerns that flowed to the White House. Anthropic, which counts Amazon as its largest investor, found itself investigated on the basis of findings from its own backer.
The company disputes the technical basis for the order. In its blog post, Anthropic said the jailbreak in question was narrow and non-universal, capable of unlocking the model’s cybersecurity abilities in a specific instance rather than defeating its safeguards wholesale. It said the same technique could elicit similar capabilities from other publicly available models, including OpenAI’s GPT-5.5, which are not subject to export controls. “We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people,” the company wrote. “If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.”
Legal scholars have raised questions about the order’s foundation. The letter reportedly cites the Export Control Reform Act’s authority for interim controls on emerging technologies, a statutory power that has never before been used as the basis for a control, as well as a regulation aimed at military-intelligence end uses that applies to a small group of adversarial countries rather than a worldwide restriction. Analysts at the Center for Strategic and International Studies concluded that the regulatory basis for a global license requirement of this kind is unclear, and that the uncertainty now extends to every company in the industry: if export controls can be applied to a model’s access, they can be applied again.
The political context sharpened the blow. In February, Trump ordered all federal agencies to stop using Anthropic’s models after the company refused contract terms that would have allowed its AI to be used for any lawful purpose, including weapons. In March, the Pentagon declared Anthropic a supply-chain risk, barring the military from using its models and prohibiting defense contractors from using them on government work. Anthropic is challenging that designation in federal court. The export control order, coming three months later, looked to many in the industry less like a safety measure than like the latest move in an escalating confrontation.
Anthropic’s response has been twofold. It dispatched senior executives to Washington to meet with White House officials and Commerce Department staff, an effort Axios described as firefighting. And it has argued publicly that the government’s own processes need reform: the company said it received only verbal evidence of the alleged jailbreak and that the letter provided no specific details of the national security concern.
The shutdown rippled through the industry and the markets. Rivals noted that a single government decision could remove an entire product line overnight, a risk investors had not priced into frontier-model companies. European customers, who rely heavily on American AI, watched with particular unease, an unease that carried directly into the G7 summit in Evian, where the episode became the reference point for every debate about AI sovereignty.
For Anthropic, the immediate question is whether access can be restored. The company has said it is working with the government on new safeguards and has signaled openness to additional technical review. The deeper question is whether the trust between a company that asked for oversight and a government that delivered it can be rebuilt. The company’s own words, in its blog post, captured the bind: “As we have stated publicly, we believe the government should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear, and grounded in technical facts. This action does not adhere to those principles.”
