Commerce Dept. Forces Anthropic Offline Over AI Security Claim

Government invokes export controls to shut down AI models

The U.S. Commerce Department sent Anthropic a letter Friday afternoon invoking an export control directive that effectively banned the company's employees and customers from accessing its Fable 5 and Mythos 5 models. The letter cited an unspecified national security concern and required immediate compliance, forcing Anthropic to shut down both flagship models to all users before the weekend.

The directive represents a swift, unilateral government action that did not appear to require court approval. Anthropic stated it believes the letter relates to a reported guardrail bypass but cannot confirm because the Commerce Department provided no specific details. The letter itself has not been made public.

The technical claim appears weak

Security researcher Katie Moussouris, founder of Luta Security, reviewed a private paper that Anthropic shared with her describing the alleged guardrail bypass in Fable 5. According to The Wall Street Journal, the paper's authors are security researchers at Amazon.

Moussouris concluded in a blog post that the bypass "should never have triggered an export control." The researchers demonstrated that slightly different phrasings—asking the model to "review code for security issues" versus "fix this code"—could produce similar results. Moussouris argued that "the behavior described in the paper cannot meaningfully be fixed, and any attempt would only weaken the model for defense."

She and dozens of other top security researchers have called on the Trump administration to revoke the order, warning that removing advanced cybersecurity capabilities from network defenders in the U.S. is "dangerous."

Political tensions, not technical risks

Axios reported over the weekend that "personality differences" between Anthropic and the Trump administration led to the export directive, rather than genuine technical concerns with the AI products. The administration has not confirmed its reasoning for invoking the export control authority.

Justin Hendrix, editor of Tech Policy Press, noted the move "is likely to raise alarms in foreign capitals about the reliability of American AI for critical applications." The message to international observers is that U.S. AI companies cannot operate without potential government interference based on political factors.

Why it matters

This incident establishes a precedent for how quickly the federal government can shut down commercial AI systems without public explanation or judicial review. The use of export controls—traditionally reserved for preventing adversaries from accessing sensitive technology—as a tool for what appears to be political pressure represents a significant expansion of executive power over the tech sector. Whether the administration misunderstood the technical details, acted on incomplete information from a competitor, or deliberately targeted Anthropic for political reasons, the result is the same: uncertainty for AI companies about what might trigger similar action and concern among international partners about the stability of U.S. technology platforms.

These details were first reported by TechCrunch and Axios.