U.S. orders Anthropic to remove two AI models over security concerns

The Trump administration ordered Anthropic to disable two of its newest AI models after receiving reports that external researchers had found ways to bypass their safety restrictions. Anthropic removed Fable 5 and Mythos 5 from availability last Friday following a letter that cited national security concerns. The letter did not detail the concerns or release an accompanying report.

Background on the order Anthropic stated it could not verify the nationality of every user and therefore chose to withdraw the models entirely. The company had previously made Fable 5 available to the public and Mythos 5 available to existing subscribers.

Reports indicated that Amazon researchers identified the bypass methods and that Amazon CEO Andy Jassy raised the issue with White House officials.

Reactions from experts and industry A group of cybersecurity researchers signed an open letter asking the administration to revoke the order. They argued that removing the models from U.S. network defenders could create security risks. Anthropic said similar bypass techniques could be applied to other AI models and that the security issues were not unique to its systems.

TechCrunch reported that some observers viewed the episode as potentially increasing interest in Anthropic's products, noting that previous disputes between the company and the administration had led to higher download numbers for its Claude model.