US Government Halts Foreign Access to Anthropic AI Model After Amazon Warning

The U.S. Commerce Department has restricted foreign access to Anthropic’s powerful Fable 5 and Mythos 5 AI models. This action, attributed to the Trump administration, follows a warning from Amazon CEO Andy Jassy regarding a potential security bypass in the Fable 5 model.

According to media reports, Jassy informed senior administration officials after Amazon researchers reportedly used prompts to extract restricted information about cyberattacks from the Mythos-class model. The exact nature of Amazon's testing—whether at the White House's request or independently—remains unclear, though an unnamed source suggested the government sought Amazon's feedback.

Concerns about unauthorized Chinese access and broader cybersecurity fears were cited as reasons for the government's intervention. While some reports indicated U.S. government suspicion of a Chinese-linked group already exploiting the bypass, an Anthropic spokesperson stated the White House did not raise Chinese access in discussions with the company, and Anthropic prohibits product access from China.

Following the warning, calls took place between Anthropic CEO Dario Amodei and administration officials. Amodei reportedly argued that the security bypass was narrow, not a full jailbreak. A source familiar with Anthropic indicated the company was given 90 minutes to disable its models without prior communication of a national security threat.

By Friday evening, the Commerce Department utilized national security export controls to prevent Anthropic from distributing Fable 5 and Mythos 5 to foreign nationals, a category including non-U.S. citizens both abroad and within the country, even Anthropic's own employees. Consequently, Anthropic disabled both models for all users.

This marks the first instance of the U.S. government using export controls to halt access to a commercial AI model already available to the public. The move has prompted international concern and intensified calls for 'sovereign AI,' advocating for national control over critical AI technology. Critics, including AI policy experts Dean Ball and Ben Murphy, have described the export controls as government overreach and warned of potential negative impacts on transparency and technological balkanization. An administration official told Axios the decision could signal increased government oversight for future AI model releases.

According to Fortune, senior technical staff from Anthropic are scheduled to meet with White House officials in Washington D.C.