
Anthropic reinstated Fable 5 globally except for sanctioned countries after the White House eased an export-control order. The deal requires real-time monitoring of the model's output for potential weapons-design queries.
Anthropic restored access to Fable 5, its flagship public AI model, after a weeks-long standoff with the Trump administration over export controls, the company said Wednesday.
The dispute began in early June when the White House, citing national security concerns around advanced AI capabilities, ordered Anthropic to disable Fable 5 globally, not just in restricted markets. The company initially complied but said the blanket suspension was “disproportionate” and launched negotiations with the Commerce Department and the National Security Council.
Those talks produced a compromise. Fable 5 is now live worldwide except in countries on the Treasury Department's sanctions list, including China, Russia, and Iran. The administration also agreed to lift a simultaneous ban on Mythos 5, Anthropic's less powerful but widely-used enterprise model, the company said.
Under the deal, Anthropic must implement real-time monitoring of Fable 5's output for any attempt to generate code or text related to nuclear, chemical, or biological weapons design. The company is also required to report any such attempts to the FBI within 24 hours, according to the joint statement from Anthropic and the White House.
The agreement does not cover Fable 5's weights or training methodology, which remain proprietary. Anthropic said the restored access applies only to the cloud-based API and the consumer chatbot interface, not to open-source distribution.
The resolution removes a cloud over Anthropic's valuation. The company was in the middle of a fundraising round that sources told Bloomberg could value it at $60 billion. Investors had flagged the export-control fight as a material risk. Anthropic did not comment on whether the round has closed.
The White House said the deal is “a model for balancing national security with AI leadership.” Anthropic CEO Dario Amodei called it “a good-faith compromise that protects both principles.”
The next risk to watch is compliance. Anthropic's monitoring system must flag potential weapon-design queries without blocking routine coding tasks. Errors in either direction could reignite tension before the 2026 election cycle.
Prepared with AlphaScala editorial tooling from the source reporting linked above. Indexable analysis may include a cited Alpha Score value. Publishing checks screen each story before release. Educational coverage, not personalized advice.