Anthropic said it has made its two most advanced artificial intelligence models, Fable 5 and Mythos 5, unavailable to users after the U.S. Department of Commerce issued an export-control directive requesting suspension of access for foreign nationals. The agency cited national security concerns, according to the company.
The Commerce Department did not provide a detailed written justification, Anthropic said, but relied on oral evidence describing a potential exploitation method that could bypass Fable 5’s defenses and surface software vulnerabilities. Anthropic disputes that a limited, non-universal security issue merits withdrawing a commercial product intended for hundreds of millions of users.
To ensure legal compliance, Anthropic temporarily disabled Fable 5 and Mythos 5 for its entire customer base, not only for foreign users. The company said other Anthropic products remain available.
Amazon Web Services (AWS), the cloud division of Amazon, confirmed that at Anthropic’s request it suspended the services in all regions.
A shift from hardware to software controls
The measure represents a notable shift in U.S. export-control policy. Until now, Washington largely targeted chips and hardware used to run AI; this is one of the first instances of restricting the foreign availability of software models themselves. Dean Ball, a former White House official, said the new rules would effectively require all non-U.S. citizens — including those physically present in the United States — to prove their citizenship to use Anthropic’s newest models.
That stipulation could create internal tensions at Anthropic: several key employees were born outside the United States, including co-founder Chris Olah and AI researcher Andrej Karpathy.
Background tensions and security arguments
Relations between Anthropic and the federal government have been strained earlier this year after the company refused requests to provide models for domestic surveillance tasks and for developing fully autonomous weapons systems. In response, the government placed Anthropic on a supplier restriction list. The current restriction comes at a time when tensions had appeared to ease.
Anthropic presented Fable 5 recently, describing it as having “Mythos-class” capabilities, and said it worked closely with authorities before bringing the model to market to reduce security risks. Independent experts have warned that models with these capabilities could accelerate complex cyberattacks if misused, particularly against sectors relying on outdated IT infrastructure such as banking.
Anthropic also noted that competitors’ products can similarly identify minor software bugs, and argued that applying this strict standard industry-wide could effectively block most new state-of-the-art models from entering the market.
Why this matters
The move signals that the U.S. government is prepared to address risks from large language models and general-purpose AI software directly through export and access controls. In the short term, the decision affects Anthropic’s customers and cloud partners; in the longer term, it may set a precedent for how governments manage international access to AI software.
Further developments to watch include any written justification from authorities, potential legal actions by Anthropic, and responses from other industry actors.
