On June 30, Anthropic announced that the U.S. Commerce Department has decided to lift export restrictions on its AI models, Fable and Mythos. This decision comes less than three weeks after the company faced an order to halt access to some of its advanced models over national security concerns.
The U.S. government has increased its oversight of AI model releases due to fears that sophisticated AI technologies could be exploited by military and intelligence sectors, especially in nations like China and Russia.
Following a directive on June 12 that suspended Anthropic's access to its premier AI models, the government later permitted the release of Mythos 5 to a select group of “trusted” U.S. organizations. In a post on social media platform X, Anthropic confirmed that they had received notification from the Department of Commerce that the export controls on both Fable 5 and Mythos 5 would be lifted, allowing them to restore access to these models starting the next day.
According to a letter from U.S. Commerce Secretary Howard Lutnick, Anthropic has committed to actively identify and mitigate any security risks associated with their models. The company is also set to collaborate closely with the U.S. government on safety protocols for both existing and future models, including reporting any malicious activities.
However, Lutnick cautioned that the government retains the authority to revisit this decision if circumstances shift or if Anthropic does not fulfill its commitments.
While Fable and Mythos share the same underlying AI technology, Fable is tailored for widespread public use, whereas Mythos includes certain safeguards. Anthropic has put in place a new mechanism to prevent a method that the government was concerned might allow for unintended exploitation or “jailbreaking” of Fable 5. This technique, while previously noted by the government, has been addressed and patched.
Experts caution that if the Mythos models are misused, they could significantly enhance cyberattack capabilities, especially in industries like banking that depend on complex, outdated technology infrastructures.
The approval for some U.S. organizations to access Mythos coincided with OpenAI's announcement that it had postponed the public release of its new GPT-5.6 at the request of the U.S. government, limiting distribution to a small group of pre-approved partners. This selective vetting process has drawn criticism regarding the lack of transparency in determining which companies qualify as “trusted.”
Increased scrutiny of AI models follows an executive order signed by former President Trump aimed at establishing a voluntary framework for AI developers to present their "frontier models" to the government before making them available to chosen partners.
Isaac Harris, the executive director of the Frontier Security Institute, remarked that the lifting of export restrictions suggests a developing standard for U.S. models. However, he expressed concern about how the government might handle similar capabilities emerging from China that are less regulated.
Notably, both Anthropic and OpenAI are in the process of confidentially filing for initial public offerings in the U.S. Anthropic's interactions with the U.S. government have been tumultuous this year. The Pentagon recently classified the company as a "supply-chain risk," limiting its contractors from using Anthropic’s technology for projects involving the military after the firm declined to provide models for mass surveillance or fully autonomous weaponry.
