Trump Administration Forces Anthropic to Pull AI Model Offline

The U.S. government last Friday forced Anthropic to withdraw its most advanced AI model from the market with roughly 90 minutes' notice, according to a Washington Post opinion column by Fareed Zakaria. The abrupt action against Mythos, Anthropic's frontier AI system, and its commercial version Fable marks the first major confrontation over who controls artificial intelligence governance in America.

The move came despite the Trump administration having signed an executive order just two and a half weeks earlier establishing a voluntary 60-day review process for frontier AI models. Before that framework could even be designed, the Pentagon declared Anthropic a "supply-chain risk" and directed federal departments to stop using its systems.

Why it matters

This incident reveals a fundamental tension in AI regulation: whether oversight will follow institutional rules or operate through improvised executive action. The precedent affects not just one company but establishes how the U.S. will govern technology that intelligence officials say can penetrate classified government systems in hours. For international partners, it demonstrates that dependence on American AI infrastructure comes with the risk of arbitrary shutdowns.

Security concerns vs. regulatory process

The administration's concerns have substance. Gen. Joshua Rudd, who leads both the National Security Agency and U.S. Cyber Command, recently testified that Mythos penetrated nearly all government classified systems in hours, not days or weeks. The Washington Post reported that Anthropic expanded access to Mythos beyond approved limits and responded slowly to concerns about user access.

But Zakaria argues these legitimate security issues actually strengthen the case for structured oversight rather than ad-hoc decisions made by whichever faction wins internal battles. Defense Secretary Pete Hegseth accused Anthropic of "betrayal" and "duplicity," while Trump has publicly mocked the company and administration officials criticized it for being "woke" and hiring Democrats.

A proposal for institutional oversight

Zakaria proposes creating an independent AI regulatory body modeled on the Federal Reserve—combining public authority with private expertise, conducting examinations and stress tests, and establishing transparent capability thresholds. Such an institution would apply rules consistently across all companies rather than targeting individual firms.

The columnist notes this approach has worked for banking, where U.S. institutions dominate globally because of strong regulation, not despite it. He suggests democracies should coordinate standards similar to how central banks coordinate financial stability.

The Trump administration's broader approach to business regulation has included taking stakes in Intel, imposing special taxes on Nvidia, and acquiring a "golden share" in U.S. Steel—actions Zakaria characterizes as personalized and unpredictable rather than rules-based.

The details were first reported by Fareed Zakaria in The Washington Post.