U.S. Reverses Landmark AI Export Ban on Anthropic’s Top Models After Two-Week Standoff

The U.S. government has abruptly reversed its own unprecedented export controls on Anthropic’s most advanced AI models, allowing the company to restore global access to Fable 5 and Mythos 5 after a two-week forced shutdown. On June 30, Commerce Secretary Howard Lutnick informed Anthropic in a letter that the controls were withdrawn and that “a license is no longer required” for export of either model. Anthropic confirmed on X that it would begin restoring access the next day, ending what Lutnick called “a more than two-week period in which the U.S. government effectively blocked access” to the systems.

The reversal caps a dramatic chapter in AI governance: the first-ever direct export control on a commercial AI model, triggered by national security fears that the models could be jailbroken to assist in cyber operations. The original order, issued by the Commerce Department’s Bureau of Industry and Security (BIS) on June 12–13, forced Anthropic to disable the models for all users worldwide, including its own foreign-national employees. The company argued it had no way to distinguish foreign nationals from domestic users in real time without violating the directive.

The original directive was unprecedented in scope. BIS ordered Anthropic to immediately suspend access to Claude Fable 5 and Claude Mythos 5 under the Export Administration Regulations (EAR), barring access “by any foreign national, whether inside or outside the United States.” Because Anthropic could not implement nationality-based access controls on the fly, it took both models offline globally — a move that Fortune described as the company being “forced to disable all access” after Commerce used national security export controls to block distribution to any foreign person.

The stated concern was a “claimed jailbreak with national security implications” involving Fable 5, according to a CSA research note. Al Jazeera reported that the Trump administration had become aware of a potential method to jailbreak the model, prompting the emergency order. Semafor added that the move was partly linked to suspicions that a China-linked group had accessed the Mythos model. Shadow Inc. noted that Amazon had flagged cybersecurity jailbreak risks related to Anthropic’s models, contributing to the government’s sudden action.

On June 26, the administration partially eased the restrictions, allowing Anthropic to release Claude Mythos 5 to some “trusted” U.S. organizations. But it took Lutnick’s letter on June 30 to fully withdraw the controls, enabling Anthropic to restore Fable 5 and Mythos 5 to all customers — domestic and foreign — without a license.

This episode represents a watershed moment for AI export controls. The initial order treated frontier AI models like other dual-use technologies under EAR, a step many analysts had predicted but never seen executed. The fact that it was targeted solely at Anthropic — not OpenAI, Google, or others — highlights the government’s willingness to act on specific security intelligence, even when it means shutting down commercial products overnight.

The global fallout was immediate. Al Jazeera reported that the ban strained relations with allies, because the order applied equally to nations that share intelligence and defense agreements — not just rivals like China and Russia. The U.S. effectively told its closest partners that they, too, could not access Anthropic’s most capable models. The lifting of controls restores those relationships but leaves a lingering question: Will the next emergency order come with more nuance?

For enterprises that rely on Anthropic’s models, the two-week shutdown was a vivid reminder of the fragility of AI supply chains. Customers who had integrated Fable 5 or Mythos 5 into production workflows suddenly found the APIs returning errors. When the partial rollback on June 26 allowed Mythos 5 for trusted U.S. organizations, it created a two-tier access system that penalized non-U.S. customers and even U.S. companies with foreign team members.

Anthropic’s own staff were affected: foreign-national employees inside the U.S. were barred from using the models they helped build. That practical impossibility — a company unable to let its own engineers test their work — was a key reason Anthropic felt compelled to shut down globally.

The Commerce Department’s decision to lift controls may not be the end of the story. Lutnick’s post on X said the government had “worked closely with Anthropic to analyze and approve Fable 5 to ensure alignment across the US Government and strengthen America’s leadership in AI.” That suggests ongoing dialogue — and possibly a framework for future emergency reviews. Policy analysts at TechPolicy.Press and the Center for Security and Emerging Technology have already flagged this as a precedent for future AI export controls, demonstrating BIS’s willingness to impose and then withdraw such restrictions.

For now, Anthropic’s customers can breathe easy. But the precedent is set: the U.S. government has both a kill switch and a willingness to use it. The next time, the reversal may not come so quickly.