Anthropic Mythos 5 Access Restored for Trusted Partners

Anthropic (P-ANTH) said on June 26 that the Commerce Department lifted a two-week export-control block on Anthropic Mythos 5, allowing the company to provide the model to designated trusted partners and expand deployment under specified safeguards.

Commerce Department Lifts Block on Mythos 5

A Commerce Department letter states that no license is required to export, reexport, or transfer the Claude Mythos 5 model to entities listed in Annex A and their foreign-national employees, as well as to Anthropic’s foreign-national employees. Commerce Secretary Howard Lutnick wrote, “I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model,” citing significant progress in talks with Anthropic.

The letter removes a licensing hurdle for the named partners, enabling Anthropic and approved organizations to bypass export reviews that had delayed access. The easing applies only to Mythos 5; Claude Fable 5 remains subject to earlier export controls and is excluded from this arrangement. Anthropic has described Fable 5 as built on the same underlying model but configured for broader enterprise and consumer use, with more conservative safeguards around high-risk areas like cybersecurity and biology.

Annex A authorizes more than 100 vetted U.S. companies and government agencies to receive Mythos 5 under this framework. The letter also records Anthropic’s commitment to collaborate with the U.S. government on protocols, standards, and future model releases, signaling an early regulatory approach to frontier AI models beyond traditional hardware export controls.

Security Concerns and Business Outlook

On June 12, the Commerce Department issued an export-control directive requiring Anthropic to suspend access to Mythos 5 and Fable 5 for foreign nationals. Anthropic said it could not reliably enforce nationality-based restrictions at scale and disabled both models globally to comply. The directive invoked licensing requirements for exports, reexports, and transfers where there was an unacceptable risk of use in or diversion to military-intelligence end uses.

U.S. government testing found Mythos identified vulnerabilities in highly sensitive classified systems within hours. Officials described the model as breaching almost all classified government systems rapidly, raising sharp security concerns that contributed to the export controls. Separate reports linked the regulatory action to discovery of a jailbreak method that could bypass Fable 5’s safety classifiers and unlock more powerful cybersecurity functions.

Parts of the National Security Agency (NSA) that had early access to Mythos through Project Glasswing lost use of the model after the directive. Some NSA red-team units that had been using the preview no longer had access. Project Glasswing, launched in April, was designed to give vetted security researchers early access to a more capable Mythos preview.

Anthropic had said it expected the models to return online in the coming days. The new authorization for vetted partners significantly expands near-term enterprise and government use compared with the prior full suspension. As part of a structured domestic restoration, Anthropic plans to implement an identity-verification policy around early July to confirm U.S. citizenship for certain users while export-control constraints remain.

The easing of restrictions on Mythos 5 allows deployment to more than 100 U.S. institutions under specified safeguards. Discussions are ongoing about releasing Fable 5, but timing remains unclear. The company’s commitment to work with the government on future protocols suggests continued oversight of safety and deployment conditions for frontier AI models.