The U.S. Department of Commerce lifted export restrictions on Anthropic’s Claude Fable 5 and Mythos 5 artificial intelligence (AI) models, allowing the company to restore global access after implementing new safeguards aimed at preventing cyber misuse.
Anthropic announced on June 30 that Commerce had notified the company that the export controls had been lifted. The decision follows several weeks of discussions between Anthropic and the Trump administration over measures designed to limit the potential misuse of the models’ cybersecurity capabilities.
“We’ll begin restoring access tomorrow, and will share an update soon,” the company said.
In a separate post, Anthropic said Claude Fable 5 would become available globally on Wednesday, July 1.
Anthropic disabled access to its most advanced AI models for all customers on June 12 to comply with an export control directive issued by the administration. On June 27, Anthropic announced that the U.S. government had allowed Mythos 5 to be redeployed to a set of U.S. organizations that operate and defend critical infrastructure.
The company said it will deploy new classifiers designed to identify and block requests involving potentially harmful cybersecurity activities while allowing legitimate uses of the models.
As Anthropic refines those safeguards over the coming weeks, some routine coding and debugging tasks will temporarily fall back to its Opus 4.8 model to reduce false positives and better distinguish legitimate requests from attempted misuse.
Anthropic also said it has begun developing a consensus framework with Amazon, Microsoft, Google, and other Glasswing partners to assess the severity of AI jailbreaks and establish industry standards for responding to them.
Project Glasswing, which launched as part of the company’s restricted rollout of the Mythos model, brings together industry partners to evaluate the model’s use in defensive cybersecurity applications.
The company also said it will expand collaboration with the U.S. government by providing pre-release access to AI models and safety measures for evaluation, sharing information on jailbreaks and misuse, and dedicating resources to joint AI safety research.