The U.S. Commerce Department has issued a directive forcing Anthropic to pull its latest artificial intelligence models, Fable 5 and Mythos 5, offline, barring non-Americans, including Anthropic’s own employees, from accessing them. This action, which Anthropic stated it did not believe was warranted by the flagged security concern, sets a precedent for government intervention in deployed AI models through export control authorities, raising alarms among cybersecurity executives and policy analysts about the reliability of American AI for critical national applications.
Defense Secretary Pete Hegseth also sought to designate Anthropic as a supply chain risk following a contract dispute with the Pentagon. This unprecedented move against a U.S. company has been challenged by Anthropic in two federal courts. The company had sought assurance that the Pentagon would not use its technology in fully autonomous weapons or for the surveillance of Americans, while Hegseth insisted the company must allow for any uses the Pentagon deemed lawful.
State Control Over Innovation
The government’s directive marks its most significant step yet to restrict access to advanced AI models. Anthropic had released Fable widely about 1 week ago, a limited version of the more advanced Mythos, access to which was already tightly restricted due to cybersecurity fears. The Commerce Department has not provided immediate comment on its action.
This enforcement letter effectively forced Anthropic to cease public access to its models just before the weekend, demonstrating that the AI industry is not immune to government interference. TechCrunch reported the move serves as a warning to the wider tech industry: comply, or the government can shut companies and their products down. The letter itself has not been made public, and Anthropic reportedly believed it was related to a bypass of the model’s guardrails, though specific details were absent. The government’s action appeared to require no court approval.
More than 100 cybersecurity experts and leaders from companies including Adobe and Nvidia submitted a letter asking the U.S. government to lift the export control directives. They urged a commitment to an open, scientific, and transparent process for handling AI risk assessments in the future. The letter acknowledged Mythos models are “quite good” at finding software flaws and weaponizing exploits but maintained they are “not uniquely good at these tasks,” noting that many signatories regularly use other foundation and open-source models for security audits and training.
Elite Interests and National Security Pretext
The directive came 10 days ago, after President Donald Trump signed an executive order to establish a framework for the federal government to vet national security risks of advanced AI systems for up to a month before their public release. This order stated that participation by AI developers would be voluntary.
Katie Moussouris, a cybersecurity veteran and researcher who founded Luta Security, stated in a blog post that the bypass described by security researchers “should never have triggered an export control.” Moussouris further asserted that “The behavior described in the paper cannot meaningfully be fixed, and any attempt would only weaken the model for defense.” She, along with dozens of other top security researchers and experts, called on the Trump administration to revoke the export control order, labeling the decision to remove advanced cybersecurity capabilities from U.S. network defenders as “dangerous.”
Amazon CEO Andy Jassy initially raised concerns about a Fable jailbreak with the White House, suggesting a deepening complexity in Amazon’s strategic partnership with Anthropic. This indicates how elite corporate interests can influence government actions that impact national technological development.
Experts Warn of National Dispossession
Justin Hendrix, editor of Tech Policy Press, warned that the Trump administration’s move is “likely to raise alarms in foreign capitals about the reliability of American AI for critical applications.” He concluded that the message being sent is that AI companies in the United States cannot be trusted to operate without interference from the U.S. government. The experts’ letter also highlighted that China’s AI models are “only months behind the best American models,” and it is probable that China’s government possesses private capabilities beyond what is publicly known, making the removal of U.S. cyber defense capabilities particularly perilous.
Joe Khawam of the Law Reform Institute indicated that the administration is likely to enforce export controls on AI models based on what they can say, rather than solely on where they are used. This suggests a potential expansion of state control into the domain of information and expression generated by AI, further eroding the autonomy of national technological development and its potential benefits for the native population.