What happened
On June 30–July 1, 2026, Commerce Secretary Howard Lutnick issued a letter to Anthropic announcing that Bureau of Industry and Security (BIS) export controls imposed on June 12, 2026 on Claude Fable 5 and Mythos 5 were fully withdrawn. A licence is no longer required for export, reexport, or in-country transfer of either model. In exchange, Anthropic agreed to: (1) proactively detect and address security risks associated with both models; (2) work with the US government on protocols and standards for Mythos, Fable, and future models; and (3) inform the US government of malicious activity detected. Anthropic began restoring global access on July 1. The original June 12 directive — triggered by an Amazon-discovered jailbreak — had forced Anthropic to suspend both models globally for all users (including employees) because nationality verification in real-time was impossible. Commerce's Center for AI Standards and Innovation validated the updated safeguards before the lift. The lift is a distinct development from the previously-covered 'partial lift' of Mythos 5 for trusted US partners on June 26.
Why it matters
This is the first case of BIS export-control powers being used to pull a publicly deployed commercial AI model off the market globally, and the first negotiated lift with binding conditions attached. It sets a concrete precedent that the US government can exercise emergency export controls to force a specific AI model offline and that restoration requires government-validated technical mitigations plus ongoing reporting obligations. Any frontier AI developer deploying models with advanced cybersecurity capabilities now faces a credible regulatory risk of sudden global suspension. The Lutnick letter's conditions — proactive risk detection, protocol collaboration, malicious-use reporting — constitute de facto ongoing compliance obligations that go beyond the original export-control framework.
Action needed
Frontier AI developers must assess whether their models could be designated 'covered frontier models' under EO 14409 and implement proactive security risk detection, government liaison protocols, and malicious-use reporting mechanisms now — not after a suspension order arrives. Enterprises depending on Anthropic models for critical workloads must build continuity plans for sudden model suspension.