Trump Administration Forces Anthropic to Suspend AI Models

The Trump administration has ordered Anthropic to suspend public access to its newest AI models, Fable 5 and Mythos 5, just days after their launch last week. The White House issued an export control directive Friday that prohibits the company from providing access to any foreign national, whether inside or outside the United States.

The unprecedented move stems from what the government claims is a cybersecurity vulnerability that allows users to "jailbreak" the models—tricking them into performing actions they're designed to prevent.

The disputed jailbreak

Anthropic has sent senior technical staff to Washington, D.C. for in-person meetings with White House officials, according to a source familiar with the situation. The company's team has also held daily virtual discussions since the directive was issued.

At the center of the dispute is the nature and severity of the alleged vulnerability. Anthropic says the government has only provided verbal evidence of what it characterizes as a "potential narrow, non-universal jailbreak." According to the company, the technique essentially involves asking the model to review a specific codebase and identify software flaws.

Non-universal jailbreaks work only in limited circumstances, unlike universal jailbreaks that enable unrestricted use of a model for potentially dangerous purposes like cyberattacks or weapons development. Anthropic contends that other leading models, including OpenAI's GPT-5.5, are vulnerable to the same method, and that the identified vulnerabilities "all appear relatively simple."

Administration pushback

David Sacks, co-chair of the President's Council of Advisers on Science and Technology, rejected Anthropic's characterization in a post on X. He stated that both the government and a "trusted partner" consider the jailbreak serious and criticized the company for not implementing a fix despite CEO Dario Amodei's public emphasis on safety.

"The Admin's hope now is that Anthropic remediates the safety issue, the export control is lifted, and Fable goes back into general release," Sacks wrote. "It is frankly bewildering that Anthropic hasn't wanted to comply with safety requests that it previously said were its highest priority."

Why it matters

This marks the first time the U.S. government has used export controls to force an AI company to suspend a model over security concerns. The precedent could reshape how frontier AI systems are evaluated and deployed, particularly as competition intensifies between companies racing to release increasingly capable models. The dispute also highlights tensions between AI developers' assessments of their own safety measures and government security evaluations—a friction point likely to grow as models become more powerful.

Details of the export control directive and Anthropic's response were first reported by AI Watch.