Anthropic has shut down access to its most advanced artificial intelligence models, including Mythos, after the Trump administration issued an unprecedented order banning all foreign citizens from its technology. The company said in a statement that the U.S. government asked it to suspend access to the Fable 5 and Mythos 5 models to all foreign nationals "whether located in or outside the United States," citing national security concerns. A U.S. official confirmed that the Commerce Department had sent the letter. The model developer has since cut off all customer access to both systems to ensure compliance.

Never before has the U.S. government taken such comprehensive measures to restrict foreign access to cutting-edge AI models developed by U.S. companies. Both the Trump and Biden administrations have restricted the export of other key technologies such as semiconductors and supercomputers, and some have discussed whether access to AI models should be blocked. However, restrictions on the software itself have triggered multiple concerns at the constitutional and commercial levels.

Anthropic said the government issued the order because it discovered that its recently released Mythos derivative Fable 5 had the possibility of "jailbreaking" (that is, bypassing security fences). The company has previously banned Fable 5 from performing cybersecurity tasks.

"We believe it is inappropriate to recall a commercial model targeting hundreds of millions of users just because of a limited potential jailbreak risk," Anthropic said in a post on its official website. "If the entire industry implements this standard, the deployment of new models by all leading model providers will basically come to a standstill."

In the name of national security, the government has implemented widespread restrictions on a series of AI models. This move may set a precedent for all mainstream AI model developers such as OpenAI, Google parent company Alphabet, and Meta. Industry leaders such as Nvidia CEO Jensen Huang and OpenAI CEO Sam Altman have previously called on the U.S. government to instead promote the global popularization of U.S. AI systems and maintain their country's leading edge.

Anthropic said it received the government order at 5:21 pm New York time on Friday. The order, issued before get off work on Friday, is contrary to previous statements by the Trump administration and a recently signed executive order that said the government would not implement a licensing system for model review.

Friday's order could also exacerbate long-standing tensions between Anthropic and parts of the Trump administration. The AI ​​developer clashed with the Pentagon earlier this year over the use of its technology for military and surveillance purposes. Following the conflict, the government designated the company a U.S. supply chain risk and ordered federal agencies to phase out its products.

Anthropic, a non-listed company that has long regarded itself as a "more responsible AI developer", released the Mythos model to a very small number of companies and institutions for the first time in April this year, warning that it has the ability to discover network security vulnerabilities and that the risk of large-scale distribution is too high.

There are signs that the limited release could have eased Anthropic's conflicts with the Trump administration: There were reports that the U.S. government was preparing to let core federal agencies use a version of Mythos.

Mythos has also pushed the Trump administration to accelerate the formulation of AI policies. The recently issued executive order proposed the implementation of a voluntary model review system and made it clear that the executive order does not constitute a mandatory licensing system.

U.S. AI developers are racing to launch the most advanced models and prove to investors that the technology can be profitable, and the latest government restrictions buck this trend. Following SpaceX's historic IPO, both OpenAI and Anthropic plan to go public as soon as this year.

In order to seize the opportunity to release cutting-edge AI models, Anthropic published a long blog earlier this month, calling for the establishment of a mechanism for the government and AI developers to jointly decide when to slow down technology research and development to avoid potential risks.

“It would be best for the world to have a choice between demonstrating or temporarily pausing” dangerous AI research and development, the company said in a blog post at the time. AI technology is developing rapidly, and its efficiency can increase human work efficiency thousands of times, or even replace human work, thus giving rise to a series of new risks.

The European Commission said it is evaluating Anthropic's statement and continues to communicate with allies about potential risks and cybersecurity issues related to advanced AI models. The European Commission added that the latest developments highlighted the need for Europe to promote technological autonomy.