US government eased restrictions on Anthropic’s most advanced AI model, Claude Mythos 5, allowing the company to grant access to more than 100 U.S. organizations, including vast corporations and government agencies.
In a letter sent to Anthropic co-founder and chief computation officer Tom Brown and obtained by WIRED, U.S. Secretary of Commerce Howard Lutnick told the AI lab that it would allow certain trusted partners access to Mythos because he “determined that appropriate safeguards were in place.” First, Semaphore reported existence of the letter.
“Anthropic has worked with the U.S. government to address risks associated with covered models. These efforts have resulted in significant progress,” Lutnick wrote.
However, the government has not agreed to a wider rollout of the model and has made no mention of the fate of Claude Fable 5, the consumer-facing version of Mythos that Anthropic released with significant additional security features. Lutnick noted in his letter that the remaining requirements set out in the original directive he sent on June 12 remain in effect.
“We have received notice from the U.S. government that Mythos 5, our strongest cybersecurity model, may be transitioned to a small group of cyber defenders and infrastructure providers,” Anthropic spokesman Eduardo Maia Silva said in a statement to WIRED. “We are working to open up access to Mythos 5 to an approved group of providers and restore their access to Mythos 5 as quickly as possible. We are pleased with this progress and continue to work with the government to expand access to Mythos 5 and make Fable 5 available for general use again.”
Anthropic is still in talks with the White House about restoring access to Fable 5, and according to a person familiar with the matter, they are expected to continue over the weekend. Both sides hope that resolving the incident will lend a hand create a lasting policy framework for future model releases, this person said.
The partial restoration comes roughly two weeks after the White House sent Anthropic an export control directive that required the company to restrict access to Mythos and Fable 5 to foreigners, including those working and living in the United States. In response, Anthropic disabled access to the models altogether. In his latest letter, Lutnick wrote that organizations approved to operate Mythos can now allow their foreign employees to access the model, and Anthropic can do the same for its own foreign employees.
The Trump administration has become concerned about Anthropic’s rollout of Mythos after learning that the company had granted access to a South Korean telecommunications company it believes has ties to China, as WIRED previously reported. Amazon and the National Security Agency also separately raised concerns to the White House that Fable 5 could be cracked, and the confluence of events convinced officials they needed to take action.
In recent weeks, Anthropic sent senior members of its cybersecurity and AI security teams to Washington, D.C. to meet with Trump administration officials. Along with Brown, the company’s talks with the U.S. Department of Commerce were led by Sarah Heck, Anthropic’s director of public policy.
Bringing Mythos 5 back online is a promising step forward for Anthropic and the White House, but the story has raised broader questions about the overall direction of U.S. AI policy, particularly the extent to which the Trump administration will seek to control future model releases. On Friday, OpenAI announced it was delaying the release of its upcoming GPT 5.6 models in response to a request from the Trump administration.