Commerce Secretary Howard Lutnick granted permission for roughly 100 companies and federal agencies to access the model, which had been offline for two weeks
NurPhoto / Getty Images
Anthropic received U.S. government clearance on Friday to restore access to its Mythos 5 artificial intelligence model for a group of roughly 100 companies and federal agencies. A letter from Commerce Secretary Howard Lutnick to Anthropic formalized the decision, which walked back a two-week-old export control directive that had compelled the company to shut down access globally.
"I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model," Lutnick wrote in the letter. Fable 5, the other Anthropic model taken offline alongside Mythos 5, was excluded from the clearance entirely.
"We received notice from the US government that Mythos 5, our strongest cybersecurity model, can be redeployed to a small group of cyber defenders and infrastructure providers," Anthropic said in a statement. "We are working to provision the approved set of providers and restore their access to Mythos 5 as quickly as possible."
A significant portion of the newly approved organizations had previously participated in Project Glasswing, Anthropic's invite-only cybersecurity initiative, which had enrolled roughly 200 firms before the shutdown, according to Reuters. Foreign nationals working at approved firms are now exempt from export licensing requirements for Mythos 5, though organizations outside the cleared group must still obtain a license to access the model.
Anthropic said it continues to work with the government to expand access to Mythos 5 and restore Fable 5 for general use.
The freeze originated with a June 12 directive in which the Trump administration prohibited Anthropic from letting foreign nationals use either model, invoking national security concerns about the possibility that the models' safety restrictions could be defeated — a technique known as jailbreaking. Because restricting access by nationality proved technically unworkable, Anthropic had little choice but to disable both models for all users globally.
Among those leading the push toward resolution was Anthropic co-founder Tom Brown, who held meetings with Lutnick and other high-ranking administration figures in the days before Friday's announcement, according to Bloomberg. Keeping CEO Dario Amodei largely out of the direct negotiations appears to have eased tensions with officials, the outlet reported. It remains unclear what specific safeguards Anthropic put in place to satisfy the government's concerns.
Friday also saw OpenAI unveil three models — GPT-5.6 Sol, Terra, and Luna — though the company held back a broad rollout after the Trump administration asked it to restrict early availability to a vetted set of partners, echoing the dynamic that had played out with Anthropic.
Join 500,000+ readers who start their day with Quartz.
By subscribing, you agree to our Terms of Service and Privacy Policy.
