The Trump administration has taken steps to limit foreign access to certain advanced artificial intelligence technologies developed in the U.S., as indicated by a report released on Friday.
According to Axios, the administration is restricting access to Anthropic’s top AI models, specifically known as Mythos 5 and Fable 5, for foreign governments, companies, and individuals. Commerce Secretary Howard Lutnick notified Anthropic’s CEO, Dario Amodei, in a letter on Friday that these two models are now subject to export controls.
Reportedly, the administration had urged Anthropic to delay the release of the models, but the company declined. This restriction isn’t just for foreign nationals in the U.S., but extends to individuals anywhere outside of the country. Under the new rules, Anthropic will need to secure a license to export, reexport, or transfer the affected AI models domestically.
Axios noted that the export controls specifically target Anthropic’s most advanced model, which is recognized as one of the leading AI technologies available. Officials within the administration explained that the Commerce Department acted after hearing reports that another company managed to breach Mythos, raising concerns about national security. These restrictions come amid widespread apprehensions that advanced AI systems could be misused by foreign entities for military, economic, or technological advantages.
In a statement, Anthropic expressed its commitment to adhering to the Export Control Directive and confirmed it would cut off access to Fable 5 and Mythos 5 in compliance. However, the company disagreed with the government’s conclusions, arguing that the alleged vulnerability only highlighted a minor issue and didn’t justify halting the commercial deployment of its models.
“We are complying with the government’s legal direction by removing access to Fable 5 and Mythos 5 for all users,” Anthropic mentioned. “However, we believe that the discovery of a limited jailbreak possibility should not necessitate a recall of commercial models that are in use by millions. If this standard were applied across the industry, it would practically freeze the rollout of all new models from all providers.”
Anthropic stated it dedicated thousands of hours to assessing Fable 5’s safety features in collaboration with government bodies and third-party organizations, both in the U.S. and the U.K., before its release. The company maintained that its testing failed to uncover a general jailbreak capable of bypassing the model’s safeguards, asserting that the vulnerabilities identified were not broad or unique to Mythos 5.
The export restrictions emerge as the U.S. government tightens controls on sensitive technologies, including advanced semiconductors and AI-related hardware, due to rising competition from China and other foreign powers.
These models represent some of Anthropic’s most sophisticated AI systems, tailored for complex tasks such as software development, scientific inquiry, cybersecurity, and advanced reasoning. Anthropic claims that Mythos 5 possesses “the most robust cybersecurity features of any model available,” while Fable 5 is designed to be a secure version for wider usage.
The new restrictions reflect increasing worries about the power of advanced AI systems. In April, Treasury Secretary Scott Bessent and Federal Reserve Chairman Jerome Powell reportedly held an emergency meeting with Wall Street leaders to discuss potential cybersecurity threats linked to Anthropic’s Mythos AI model.
The Department of Commerce did not respond promptly to requests for comments.
