The US government has moved to restrict export of Anthropic’s most advanced artificial intelligence models, citing concerns they could be exploited by foreign military and intelligence organizations.
The move marks one of Washington’s first attempts to use export-control powers to limit access to frontier AI, underscoring the tech’s growing strategic value in military and cyber operations.
Commerce Secretary Howard Lutnick ordered Anthropic to halt supply of its Mythos and Fable models to foreign nationals as officials raised concerns that safeguards built into the systems could be bypassed.
According to the company, the government identified a potential way to “jailbreak” restrictions in its Fable 5 model, which could enable users to identify software vulnerabilities despite controls designed to prevent that activity.
Following the order, Anthropic disabled access to the affected models globally and entered discussions with the Commerce Department over conditions for restoring service.
‘Only Minor’
Reuters reported that company representatives have been meeting regularly with US officials as both sides work toward a resolution that addresses national security concerns while allowing access to resume.
While it complied with the order, Anthropic rejected the overall assessment, arguing that the issue revealed only minor security weaknesses and that similar capabilities are already available through other public AI models.
The Commerce Department is invoking powers under the 2018 Export Control Reform Act, extending authorities traditionally used for defense technologies and semiconductors into the frontier AI space.
Some export-control experts, however, question whether existing regulations apply to AI models delivered through cloud services, suggesting that remote access differs from the conventional export of controlled technologies.
The latest dispute comes amid broader tensions between Anthropic and President Donald Trump’s administration over military applications of artificial intelligence.
Earlier this year, the company opposed the use of its AI systems for domestic surveillance and fully autonomous weapons, fueling wider debates over safeguards for military AI deployments.
