US government forces Anthropic to pull AI models offline

On Friday afternoon, the U.S. Commerce Department sent Anthropic an enforcement letter invoking an export control directive—defined as a regulatory mechanism restricting access to technology. This unilateral action forced the company to pull its Fable 5 and Mythos 5 AI models offline, citing unspecified national security concerns. Anthropic believes the letter is related to a bypass of the models’ guardrails—the safety mechanisms in AI models—but the company isn’t sure of the exact cause because the government’s letter lacks specific details. In response to the directive, Anthropic shut down both models to all customers to ensure compliance. The sudden shutdown highlights the immediate operational impact of government intervention on technology firms.

The Trump administration’s directive has not been fully explained, leading to criticism over the government’s shaky reasoning. Observers note that the administration’s directive appears retaliatory rather than being based on a clear technical issue. This action marks a significant escalation in government intervention, demonstrating how unilateral government actions can abruptly force a technology company to take its products offline. The precedent recalls past export laws from the 2010s, showing that the AI industry is increasingly subject to direct political and regulatory pressures from the U.S. government.

Cybersecurity experts have raised concerns about the broader implications of the decision. Katie Moussouris, a cybersecurity researcher and founder of Luta Security, characterized the move to pull advanced cybersecurity capabilities as dangerous. Regarding the security paper that prompted the discussion, Moussouris stated, “The behavior described in the paper cannot meaningfully be fixed, and any attempt would only weaken the model for defense.” This warning highlights the tension between national security directives and practical cybersecurity defense.

Additionally, Justin Hendrix, the editor of Tech Policy Press, stated that the move is likely to raise alarms in foreign capitals about the reliability of American AI for critical applications. Hendrix also noted a cloud of suspicion that senior officials are picking favorites based on personal and political factors. This development suggests that AI companies in the United States may face regulatory actions, complicating their relationships with foreign partners who rely on American technology.

Why it matters

The U.S. government’s unilateral action sets a precedent for government control over software, raising significant concerns about the reliability of American AI for critical foreign applications.