Trump administration gives Anthropic 90 minutes to pull AI models

The Trump administration gave Anthropic 90 minutes to take down its newest AI models, turning a fast-moving safety dispute into a test of how far federal emergency power can reach into frontier software. The order, delivered at about 5:21 p.m. ET on June 12, forced Anthropic to disable access to Fable 5 and Mythos 5 and raised an immediate question: what legal authority lets Washington compel a private AI company to pull a model offline on that timeline?

Anthropic said the directive came from the U.S. government citing national security authorities and ordered access suspended for any foreign national, whether inside or outside the United States, including foreign-national Anthropic employees. Because the company said it could not separate foreign users from domestic customers in real time, it shut off access for all users to comply. The models had launched only days earlier, making the action one of the sharpest regulatory interventions yet in the commercial AI market.

The 90-minute deadline followed a frantic 24-hour effort by senior administration officials to persuade Anthropic to pull the model voluntarily before the government moved to a licensing regime, according to reports. Amazon CEO Andy Jassy was among the tech leaders who raised security concerns with senior Trump administration officials that week, and those warnings helped accelerate the response. Anthropic said Fable 5 was its broadly released model, while Mythos 5 carried extra safeguards for cybersecurity and biology, routing those queries to Opus 4.8.

The episode is now being read in Washington as a precedent-setting moment for AI policy. The White House had already pushed an AI agenda centered on innovation and security, and the Anthropic order showed how quickly export controls can be used not just on chips or hardware, but on model access itself. Industry observers have described it as the most significant U.S. step yet to restrict access to frontier AI software, and as a likely early test of whether national-security tools designed for goods crossing borders can be adapted to software that can be switched off with a call.

For Anthropic, the conflict landed in a difficult place: between government pressure, customer expectations and the practical limits of model access controls. The company said it disagreed with the government’s handling, while customers began asking about refunds tied to Fable 5 purchases. However the dispute is resolved, the larger issue remains open. The federal government has now shown it can move from concern to compulsion in less than two hours, and the AI industry will have to reckon with what that means for due process, competition and speech.