US lifts export controls on Anthropic's Fable 5 and Mythos 5 AI models after three-week shutdown

The shutdown

On June 9, Anthropic released Fable 5, a restricted version of its powerful Mythos 5 model, with safeguards against cybersecurity misuse. Three days later, the US Commerce Department issued an export control directive that forced Anthropic to disable both models for all users worldwide, including foreign employees. The company described the move as based on "a misunderstanding" around possible jailbreaks.

“We have stated publicly, we believe the government should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear, and grounded in technical facts. This action does not adhere to those principles.”

A partial reversal

On June 27, the administration allowed Mythos 5 to be restored for a limited set of US organizations, leaving foreign governments, companies, and financial institutions without access. Fable 5 remained fully blocked, creating what the Wall Street Journal called a situation "sans précédent pour une entreprise d'IA américaine de premier plan", unprecedented for a leading American AI company.

The agreement

Commerce Secretary Howard Lutnick announced on June 30 that an agreement had been reached. Under the terms, Anthropic committed to proactively detecting and addressing security risks, working with the US government on protocols and standards for current and future models, and reporting malicious activity. The export controls were then lifted.

“We received notification that the Department of Commerce has removed the export controls for Claude Fable 5 and Mythos 5. We will begin restoring access tomorrow.”

Restoration and remaining limits

Access to Fable 5 is being reestablished from July 1 across platforms including AWS, Google Cloud, and Microsoft Foundry. Mythos 5, however, remains restricted to a narrow group of pre-vetted US partners, and Anthropic said it would continue coordinating with the government to widen that circle under its Glasswing program. The partial lifting leaves non-US cybersecurity agencies in Europe and Asia still waiting.

What it means for AI regulation

The episode marks the first time a major US AI release has been halted and then negotiated back by the government. It signals that the most capable models are increasingly treated as strategic technologies, not routine software updates, and may reshape how future launches are reviewed and controlled.