Anthropic’s AI Blocked by U.S. Export Order, Exposing a New Security Fight

Anthropic unveiled its most powerful AI model last Tuesday, then saw it blocked by Friday evening at 5:21 p.m. Eastern time after the U.S. Commerce Department issued an export-control order. The order barred access to the company’s new Claude Fable 5 model and the stronger Mythos 5 version for any foreign national in the United States or abroad, including non-American Anthropic employees, forcing Anthropic to disable both models for all customers.

Anthropic says the public Fable 5 model includes built-in safeguards meant to prevent misuse in areas such as biology and chemistry, while Mythos 5 is the uncensored version available only to vetted cyber and infrastructure partners. The government said it had learned of a way to bypass Fable 5’s protections, but Anthropic rejected that characterization, saying the issue was a normal request to read code and fix bugs, a capability it says exists in rival systems too. The company called the dispute an “understanding” and said it would release more details.

The episode has no clear precedent, and both sides are improvising in real time. Anthropic says it wants governments to have power to block unsafe models, but only through a transparent and fair process. The Commerce Department is using export-control rules built for chips, weapons, and physical technologies, even though these models exist in the cloud and can be accessed globally.

The clash comes amid months of tension between Anthropic and the Trump administration. In February, Trump and Defense Secretary Pete Hegseth reportedly barred Anthropic products in federal agencies after the company sought tighter limits on Pentagon use of its technology. Trump called Anthropic staff “crazy leftists” who made a “disastrous mistake.” Anthropic sued, won a ruling in California, but the case remains active in Washington.

The story deepened after the Financial Times reported that the U.S. National Security Agency had used Mythos for offensive cyberattacks. A source cited by Axios later said the government tried to delay Anthropic’s launch, failed, and only then sent the export-control letter. The same source said the model must stay locked until U.S. national-security capabilities are strengthened, possibly within weeks. What happens next, in court, in talks, or through public pressure, could set the default for future AI regulation and split advanced AI access by nationality and geopolitics.