The U.S. government has removed the licensing requirement for Anthropic, allowing the company to export its advanced AI models, Mythos and Fable, which had limited public access.
Anthropic announced that it would start restoring accessibility to these models on Wednesday, July 1.
Previously, on June 12, the U.S. classified these technologies as export-control items, necessitating special permission for any distribution to foreign entities. This made it impractical for Anthropic to sustain public access, leading to a complete halt in sharing the models.
After extensive negotiations, Commerce Secretary Howard Lutnick stated that Anthropic has committed to identifying and mitigating security risks tied to its models. The company will collaborate with the U.S. government on necessary protocols and report any malicious uses.
Anthropic had already signaled its intent to undertake many of these measures voluntarily, prior to the implementation of the export controls. This fueled skepticism among cybersecurity experts regarding the necessity of the restrictions, with some viewing them as a political maneuver against Anthropic due to the critical stance of its executives toward governmental practices and potential abuses of technology.
Mythos was primarily released to a limited range of organizations starting in April, aimed at addressing fears over its potential to find and exploit software vulnerabilities. Meanwhile, Fable was made available to the general public in June with added safety features.
With competitors in Asia starting to launch AI models that rival the capabilities of Mythos—such as Fugu and Tulonfeng—the U.S. faced mounting pressure to relax its restrictions on Anthropic to maintain global competitiveness in AI technology.
Last week, Lutnick permitted the release of Mythos to select clients authorized by the White House. OpenAI’s recent models also followed a similar path, being distributed to a chosen set of organizations rather than released publicly.
The Trump administration’s inconsistent approach to AI policy has created uncertainty for companies regarding future frameworks for releasing models. An executive order from June, suggesting a need for pre-release evaluations, drew criticism from notable analysts like Dean W. Ball, who recently joined OpenAI in a policy role.
