The US government has sparked concern across the technology sector after its Commerce Department issued an enforcement letter to artificial intelligence firm Anthropic, compelling the company to withdraw its advanced cybersecurity models, Fable 5 and Mythos 5, from public access. The directive, issued last Friday, cited an unspecified national security concern under an obscure export control law, banning non-Americans, including Anthropic's own employees, from accessing the sophisticated AI systems.
Anthropic responded by immediately shutting down both models to all customers, ensuring compliance with the government's order. While the company suggested the letter might relate to a bypass of the models' safety guardrails, it acknowledged a lack of specific details from the government. The letter itself has not been made public, leading to widespread speculation regarding the precise nature of the alleged threat.
However, new information emerging over the weekend has cast doubt on the official reasoning. Katie Moussouris, a cybersecurity veteran and founder of Luta Security, revealed in a blog post that Anthropic had shared a private paper with her detailing an alleged guardrail bypass in Fable 5. Moussouris, who was asked for her assessment, concluded that while a bypass was described, it should not have triggered an export control. She highlighted that the distinction between asking an AI to 'review code for security issues' versus 'fix this code' is largely semantic, with similar end results for cybersecurity defence.
Moussouris criticised the export control directive as hasty and misguided, arguing that the described behaviour 'cannot meaningfully be fixed' and any attempt to do so would only weaken the model's defensive capabilities. She, along with dozens of other leading security researchers, has called on the Trump administration to revoke the order, deeming the removal of advanced cybersecurity tools from US network defenders as 'dangerous'.
The incident has been interpreted by some as a potential warning to the wider tech industry, signalling that AI companies are not immune to unilateral government intervention. Justin Hendrix, editor of Tech Policy Press, suggested that the move is 'likely to raise alarms in foreign capitals about the reliability of American AI for critical applications,' implying that US AI firms may not be trusted to operate without government interference. The lack of transparency from the Trump administration on the specific reasons for the ban continues to fuel speculation about whether the decision was technically justified or driven by other factors.