The Trump administration has partially lifted its ban on Anthropic's Claude Mythos 5 AI model, allowing more than 100 U.S. government agencies and companies to access it. This decision comes two weeks after an initial ban was imposed due to national security concerns, specifically that a group linked to China had accessed Mythos. Commerce Secretary Howard Lutnick confirmed the eased restrictions, noting that Anthropic has committed to working with the U.S. government on protocols and standards for future releases of its models.
Anthropic publicly acknowledged the development, stating they've been working closely with the U.S. government since June 12 to restore access to Mythos 5 and Fable 5. Mythos 5, their strongest cybersecurity model, can now be redeployed to a set of U.S. organizations that operate and defend critical infrastructure. They are continuing discussions to expand access to Mythos 5 and make Fable 5 available for general use again. Fable 5, a version of Mythos 5 designed for general public release with more relaxed guardrails, remains restricted.
The initial ban on Mythos 5 and Fable 5, imposed on June 12, led Anthropic to take both models offline worldwide and limit access for foreign nationals. The government's decision to ease restrictions on Mythos 5 now also permits non-American employees of the approved organizations, as well as Anthropic's own non-American staff, to access the model. However, security experts have warned that Mythos models, if misused, could significantly enhance cyberattacks on critical infrastructure sectors. The Commerce Department indicated that appropriate safeguards are now in place, although specific details of these safeguards remain opaque.
The selection process for the more than 100 approved organizations has drawn criticism for its lack of transparency. Analysts and rights groups, including John Coleman of the Foundation for Individual Rights and Expression, have questioned the opacity of the vetting process, arguing it concentrates too much power with the government and lacks accountability. OpenAI CEO Sam Altman also voiced unease about the government's role in picking customers for AI models. Despite the partial reprieve, the broader issue of how companies can widely release updated AI models amidst tightening regulatory scrutiny remains unresolved. This move follows an executive order by President Trump establishing a voluntary framework for reviewing frontier AI models before release.
