Anthropic models pulled after US export-control letter

The U.S. Commerce Department sent Anthropic a letter on Friday afternoon invoking an obscure export control directive that barred non-U.S. persons, including Anthropic employees, from accessing Fable 5 and Mythos 5. Anthropic said it believes the letter is related to an alleged bypass of the models’ guardrails, but the department did not provide specific details and the letter has not been made public.

Anthropic responded by shutting down both of its top models to all customers to ensure compliance with the directive. The action effectively removed those models from use while the company and officials sorted the issue.

What happened and how it unfolded

Commerce Department officials cited an unspecified national security concern when they told Anthropic that non-Americans could not access Fable 5 and Mythos 5. The enforcement letter, as described in the coverage, forced the company to act quickly and unilaterally to avoid violating the order.

Axios described tensions over the weekend between Anthropic and the administration, saying “personality differences” between the company and the Trump administration contributed to the export directive rather than a purely technical issue with the models. The Wall Street Journal reported that the security paper describing the alleged guardrail bypass was authored by security researchers at Amazon.

Anthropic shared a private copy of that paper with Katie Moussouris of Luta Security and asked for her assessment. Moussouris wrote in a blog post that the researchers demonstrated a guardrail bypass in Fable 5 but argued the bypass “should never have triggered an export control.” She also wrote that the bypass “cannot meaningfully be fixed, and any attempt would only weaken the model for defense.”

Dozens of security researchers and experts have since urged the administration to revoke the export control order, arguing the directive removes advanced cybersecurity capabilities from network defenders in the United States and is therefore dangerous.

Prior context from past policy moves

Past administrations created broad language in export rules covering cybersecurity tools, and that vagueness nearly outlawed legitimate security and vulnerability research during the 2010s, the coverage notes. The current directive appears to revive similar concerns about sweeping language and collateral effects on legitimate security work.

Observers have framed the most recent action as partly political. Justin Hendrix, editor of Tech Policy Press, said the move is “likely to raise alarms in foreign capitals about the reliability of American AI for critical applications.” Commentators suggested the directive may reflect friction between Anthropic and government officials rather than a clear-cut technical violation, and that the White House might not have anticipated the breadth of its effects.

Why it matters

The government’s ability to force an American AI company to take models offline with a unilateral export-control demand changes the relationship between AI labs and regulators. Companies that build or host models now face a precedent where a single enforcement letter can disrupt service and remove capabilities from users without a public explanation or court process.

That matters for defenders and customers. Security researchers argue the order stripped network defenders of advanced tools; others worry the action will sow distrust abroad about whether U.S. AI providers can deliver reliable services free from sudden political intervention. The episode also highlights the tension between fixing models to prevent misuse and preserving their utility for legitimate security work.

What to watch

Whether the administration makes the Commerce Department letter public and whether it revokes or clarifies the order are the immediate signals to monitor. Also watch for any similar enforcement actions against other U.S. AI companies and for follow-up reporting on the guardrail bypass paper whose authors the Wall Street Journal identified as Amazon security researchers.

If officials publish the rationale or withdraw the directive, that would reduce uncertainty; if not, other firms and foreign customers may reassess reliance on U.S.-based models.