Anthropic halts its most powerful AI models after US export control order

Anthropic has taken its two most advanced artificial intelligence models offline, just days after their launch, to comply with a US government directive that bans access for all foreign nationals. The company received the order from commerce secretary Howard Lutnick on Friday afternoon, based on export control regulations.

The government directive

The order prohibits access to Fable 5 and Mythos 5 for 'any foreign citizen, inside or outside the United States', including Anthropic's own foreign employees. Anthropic said it could not filter users by nationality and had to disable the models for everyone to ensure compliance. The company described the move as an 'abrupt' step and apologised to customers.

The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Anthropic models will not be affected.

The jailbreak concern

According to the company's statement, the US government believes it has discovered a method to bypass, or 'jailbreak', Fable 5. Anthropic said it was shown only limited evidence and considers the action a 'misunderstanding'. The company is working to restore access 'as soon as possible'.

We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people.

Fable 5 and Mythos 5

Fable 5, launched widely on Tuesday, is a restricted version of the more powerful Mythos 5. Mythos 5 had already been available only to a select group of organisations because of its advanced cybersecurity capabilities. The model can identify software vulnerabilities, some unknown for decades, that could be exploited in attacks on critical infrastructure.

Escalating tensions

The suspension is the latest clash between Anthropic and the Trump administration. In February, the president ordered federal agencies to phase out Anthropic technology after a dispute with the Pentagon. A federal judge later blocked the Pentagon from designating the company as a supply chain risk. Ten days before the Friday directive, Trump signed an executive order that stopped short of giving the government power to block model releases, setting up only a voluntary vetting framework.

The confrontation threatens Anthropic's plans for a Wall Street listing, possibly later this year, at a valuation near $1 trillion. Investors have grown concerned about regulatory risk and the company's ability to maintain its technology lead.