The U.S. government lifted its export suspension on Anthropic’s Claude Mythos 5 on Friday. The decision makes the model available to more than 100 institutions in the United States, including major companies and government agencies.
This action is a reversal after the Trump administration and Anthropic faced off for two weeks. Mythos 5 was approved, but Fable 5, the consumer version, remains unavailable
The U.S. Department of Commerce cleared Claude to Mythos 5 for trusted partners
Commerce Secretary Howard Lutnick said in a letter to Tom Brown, the chief of computing at Anthropic, on Friday that there is no longer any need to obtain a license for the export of Mythos 5 to the agencies listed in Appendix A
I have reviewed and determined that appropriate protective measures have been put in place to allow some trusted partners to access the Claude Mythos 5 Model. Commerce Secretary Howard Lutnick, Semafor
Follow us on X for the latest updates as events unfold
Senior officials from Anthropic traveled to Washington to meet with administration officials during this dispute, according to a CNBC report
This change means that the models behind the launches of Mythos and Fable have escaped the controls that were just set earlier this month. Before that, both models had been temporarily disabled after Amazon, one of Anthropic’s largest investors, issued a warning. Researchers at the company cautioned that Fable 5 could be jailbroken to be used in a harmful way
Before that, Mythos was in the Glasswing program, a vulnerability search initiative involving around 150 organizations across more than 15 countries. The model detected secret system flaws within just a few hours after the test began with the government
Fable 5 is still waiting under the new era of AI
Sources close to the matter say that the rollout of Fable 5 is underway, even though the timeline is still unclear. Unlike Mythos, Fable 5 previously was available to anyone with a membership subscription. Fable 5 had been the most powerful AI tool, allowing the public to try it out
The event is becoming a new screening system. On June 2, an executive order established a voluntary channel for government agencies to test frontier models. Developers could submit models for cybersecurity review within 30 days before they are released to the market. Over the past year, Washington has tightened AI chip exports to China even more, and extending that power to control access to models is another front
OpenAI followed suit on Friday, limiting access to its most powerful GPT-5.6 model, named Sol, to only about 20 government-approved partners. Lower-capability versions, Terra and Luna, have been opened for public use
This restriction began out of concerns about China’s access, with a report linking the issue to SK Telecom, South Korea’s communications company, which was incorporated into Glasswing in early June, before access to SK Telecom was shut down. SK Telecom denied that the company had any links to China
Dozens of cybersecurity leaders have banded together to urge the administration to rescind those control measures. A sealed letter initiated by Alex Stamos, the former head of security at Facebook, has gained support signatures from leading companies such as Nvidia, Adobe, and Zoom
As European and global partners grow frustrated at having to rely on Washington for access, if Fable 5 is approved on equal terms, clarity may emerge in the next few days
