Trump Administration Bans Foreign Access to Anthropic's Most Powerful AI Models

The Trump administration delivered a stunning regulatory blow to Anthropic on Friday, with Commerce Secretary Howard Lutnick sending CEO Dario Amodei a letter declaring the company's two most advanced models — Fable 5 and Mythos 5 — subject to export restrictions. The ban prohibits usage by foreign governments, companies, individuals, and even foreign nationals living inside the United States. The scope was so broad that Anthropic shut down access to both models entirely, since the restriction sweeps in many of its own foreign-born employees.

The immediate practical consequences are significant. Governments and enterprises worldwide had been relying on Mythos 5 to identify and patch software vulnerabilities, and that work has now halted overnight. Critical sectors — finance, energy, and beyond — had already deployed the model. Fable 5, billed as a next-generation version of Mythos with safety guardrails built in, had only been released to the public days earlier. Anthropic said its other products remain available to customers unaffected.

The trigger appears to be a jailbreak demonstration conducted by Amazon researchers, who used a series of specially crafted prompts to extract information about a small number of known software vulnerabilities from the model. Cybersecurity CEO Katie Moussouris of Luta Security, who reviewed the report Anthropic shared with her, said the information would be far more useful to network defenders than attackers, calling the government's response a complete overreaction. Amazon did not respond to requests for comment.

The action escalates months of simmering tension between Anthropic and the White House. Administration officials and advisers — including venture capitalist David Sacks — have repeatedly accused Anthropic of so-called doomerism, arguing the company stokes fear about AI risks while continuing to release increasingly powerful tools. A recent executive order signed by President Trump gave national-security and cybersecurity officials greater say over model evaluations, and the two sides have been negotiating over access to Anthropic's top tools for weeks.

Analysts noted a glaring contradiction in the administration's stance: while forcefully restricting access to Anthropic's models on security grounds, the same administration has loosened chip export rules and even approved semiconductor shipments to China. Jimmy Goodrich, a senior fellow at the UC Institute on Global Conflict and Cooperation, flagged the inconsistency directly. The move is considered unprecedented — no U.S. government has previously deployed such punitive export controls against one of its own leading AI companies — and arrives as Anthropic and rivals are racing toward highly anticipated IPOs, making any regulatory disruption potentially worth billions.