Anthropic cuts off access to AI models after U.S. export order

Anthropic cut off worldwide access to two of its most advanced AI models after the Trump administration ordered the company to block foreign nationals, whether inside or outside the United States, from using them. The decision puts a private AI safety dispute squarely inside federal export policy, and it shows how pressure from senior tech executives, including Amazon CEO Andy Jassy, can help shape both market access and regulatory action.

Anthropic said it disabled access to Fable 5 and Mythos 5 globally to comply with the export-control directive. The company said the government believes there is a way to bypass, or jailbreak, a safeguard meant to stop Fable 5 from being used to identify software vulnerabilities. Anthropic said its own review of the demonstration found only a small number of previously known, minor vulnerabilities, and added that other publicly available models could find the same issues without requiring a bypass.

That pushed the dispute beyond a normal product-safety debate. Anthropic said the government’s action did not meet a transparent, fair, clear and technically grounded safety process. The move also marked a new kind of AI restriction, one that goes beyond chips and hardware to directly limit access to advanced models, including among allied-country users.

The company had already been tightening access to these systems. In April, Anthropic first made Mythos available to about 40 organizations that maintain critical computer infrastructure, then expanded that pool to 150 organizations. On Tuesday, it unveiled Fable 5 as a more tightly guarded version of Mythos, designed with additional guardrails to reduce cybersecurity risks.

The policy stakes are sharper because Anthropic is also navigating corporate and political pressure at the same time. The company has confidentially filed for a U.S. initial public offering. It has also faced earlier tensions with the Trump administration, even as those frictions had reportedly begun to ease. Anthropic chief executive Dario Amodei had supported the president’s broader call for government oversight of new AI systems just days earlier.

Not everyone in the national-security world saw the order as carefully calibrated. Jimmy Goodrich of the University of California’s Institute for Global Conflict and Cooperation called the move “not well thought-out,” saying it also bars Canadians and Brits employed at Anthropic from research and development. The episode adds to a broader federal pattern: earlier in 2026, Defense Secretary Pete Hegseth labeled Anthropic a supply chain risk and barred use of its Claude AI chat model after a separate dispute over safeguards.