Anthropic Races to Washington After Trump Export Ban Forces Its New AI Model Offline

Anthropic dispatched a senior technical delegation to Washington on Monday for face-to-face meetings with Trump administration officials, scrambling to lift a federally imposed export ban that forced the AI startup to pull its newest model, Fable 5, from global access Friday night. The ban — the first time the White House has compelled a company to remove an AI model from public availability — bars foreign nationals from using Fable 5, citing potential security vulnerabilities discovered by Amazon, a major Anthropic investor.

The crisis began when Amazon researchers found a way to bypass Fable 5's built-in guardrails and reported the finding to the White House and the National Security Agency. The discovery alarmed officials including National Cyber Director Sean Cairncross and Treasury Secretary Scott Bessent, who demanded Anthropic halt access to the model. When Anthropic pushed back — arguing the breach was minor and not a full 'jailbreak' enabling cyberattacks — the administration imposed the export control, effectively forcing a total takedown since Anthropic couldn't easily segregate U.S. from non-U.S. users.

The weekend featured marathon calls between Anthropic co-founder Tom Brown, Commerce Secretary Howard Lutnick, and Cairncross. Monday's more technical in-person sessions at Commerce and Cairncross's office were staffed by Anthropic's Logan Graham (Frontier Red Team), Dave Orr (head of safeguards), and Nicholas Carlini (lead security researcher), alongside Commerce's Chris Fall, who heads the Center for AI Standards and Innovation. Anthropic presented its cybersecurity protocols in hopes of clearing the path back to market.

A resolution is not imminent. A senior White House official signaled the timeline is squarely in Anthropic's hands, while warning that a prolonged standoff risks metastasizing into a de facto AI licensing regime — one where companies must seek government permission before releasing new models. One current administration official called that prospect 'an extremely bad situation' that would hobble U.S. competitiveness against China. A letter signed by nearly 80 tech leaders and CEOs backed Anthropic, arguing the capability Amazon flagged is routine for any model designed to write secure code and poses no genuine offensive risk.

The episode has exposed a deep fault line: Fable 5 was intentionally less powerful than Anthropic's elite Mythos model — itself restricted to a select few companies due to its ability to exploit cyber vulnerabilities faster than any human — but performed at a comparable level while including safeguards against bioweapon or cyberattack prompting. Critics within the administration itself warn that improvised, opaque regulatory moves are creating a chilling uncertainty across the entire AI industry.