Trump administration bars foreign access to Anthropic's Mythos and Fable models

The Trump White House has imposed strict export controls on Anthropic's advanced AI models, Mythos and Fable, forcing the company to take them offline after reports of potential jailbreaks raised national security alarms. The move, prompted by a call from Amazon CEO Andy Jassy to Treasury Secretary Scott Bessent, bars foreign nationals from accessing the systems, including Anthropic's own non-U.S. employees. Administration officials framed the decision as a necessary safeguard against threats that could undermine American interests.

Yet the episode has exposed deep tensions between the company and the administration, with officials accusing Anthropic of failing to communicate in terms the president understands. Sources close to the matter described clashes over how seriously the firm treated warnings about vulnerabilities in Fable, which researchers reportedly bypassed in ways that could expose sensitive capabilities. Anthropic countered that the issue was a narrow flaw not unique to its models and that it had received prior government approval to deploy Fable. The company ultimately pulled both models from public access rather than comply with the restrictions.

White House officials expressed frustration that Anthropic had spent months warning about catastrophic AI risks while appearing to downplay its own security lapses. One senior administration figure told reporters the episode showed a self-serving approach, noting that nearly half a dozen other companies had flagged concerns before action was taken. The rhetoric reflects an administration eager to project strength on technology controls, even as the abrupt ban disrupts a leading U.S. AI developer.

The timing has drawn particular scrutiny. Just days after the curbs took effect, shares in Chinese AI firm Zhipu surged more than 30 percent, with Wall Street analysts upgrading the company and citing new opportunities to fill gaps left by restricted American models. Zhipu quickly announced plans to release its own advanced GLM-5.2 model as open-source software without usage limits. Rival MiniMax also saw gains. Critics argue the policy risks accelerating the very technological competition with China that the administration claims to prioritize.

Anthropic's CEO Dario Amodei, long vocal about AI dangers, now faces investor concerns over the company's handling of the dispute. The firm maintains the White House overreacted to a manageable issue, yet the decision to shutter the models entirely underscores how quickly regulatory pressure can sideline even domestic leaders in the field. Observers note that personality-driven friction appears to have shaped the outcome more than coordinated strategy.

The broader fallout highlights inconsistencies in the administration's approach. While officials tout export controls as essential protection, the immediate beneficiary has been a foreign competitor unburdened by similar restraints. Anthropic's experience suggests that heavy-handed tactics without clear international coordination may weaken rather than strengthen U.S. positioning in AI development.