The Commerce Department on Friday partially lifted its two-week block on Anthropic’s Mythos 5, clearing the company’s strongest cybersecurity model for roughly 100 vetted companies and federal agencies. In a letter to Anthropic co-founder Tom Brown, viewed by CNBC, Secretary Howard Lutnick wrote that “appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model.” Hours later, OpenAI announced it would roll out its GPT-5.6 series to a similarly narrow circle of administration-approved partners. The two announcements, landing the same day, mark something that hadn’t really existed in American AI policy before: a federal gatekeeping function over which firms can touch frontier models.

The original shutdown earlier in June came via an export-control directive citing national security authorities. Because foreign nationals work at Anthropic and at many of its partner organizations, the framework forced both Mythos 5 and the consumer-facing Fable 5 offline entirely, according to NBC News. Anthropic said the model is now being “redeployed to a small group of cyber defenders and infrastructure providers.” Before being pulled, Mythos 5 had run inside Project Glasswing at organizations including Cisco and JPMorgan Chase, where an early version had reportedly identified thousands of previously unknown cybersecurity vulnerabilities.

That number is the policy story. A model capable of surfacing thousands of latent exploits is, by definition, dual-use, and the export-control posture treats it accordingly. Lutnick told Bloomberg that Anthropic had “worked with the US government to address risks associated with the Covered Models” and that “these efforts have yielded significant progress.” Behind the language, Anthropic dispatched senior scientists and engineers to Washington to work with Commerce and the Office of the National Cyber Director. Fable 5 remains offline; CNN reported that talks aimed at restoring it were expected to continue into the weekend.

Semafor, which first reported the Mythos clearance, characterized the Lutnick letter as the opening of a new regulatory regime in which Washington controls frontier-model release. European officials and other U.S. allies have already registered frustration at their new dependence on those decisions.

“In just two weeks, we have worked diligently to ensure America remains the global leader in AI while safeguarding our security,” said Benno Kass, a Commerce Department spokesman. It’s a sentence that reads as procedural and is, in fact, constitutional: the federal government now decides who gets the model.

Sources