U.S. allows limited return of Anthropic’s Mythos 5 AI model

The Trump administration lifted its block on Anthropic’s Mythos 5 for roughly 100 U.S. companies and federal agencies, ending a June 12 export-control order that had forced the company to cut off both Mythos 5 and Fable 5 to any foreign national. The reversal is a partial one: the model is back only for a limited set of approved users, and the government is still treating frontier AI access as a national-security decision, not a normal product launch.

Commerce Secretary Howard Lutnick said in a letter that he determined “appropriate safeguards are in place” to let “trusted partners” use Claude Mythos 5. The revised licensing terms reportedly remove the export-license requirement for the entities named in the annex, along with their foreign-national employees and Anthropic’s foreign-national employees. That change marks the sharpest shift in the government’s risk calculus: after ordering a broad shutdown, officials are now carving out a controlled channel for institutions they believe can use the model without worsening the security risk.

Anthropic said the government notified it that Mythos 5 could be redeployed to U.S. organizations that operate and defend critical infrastructure, and that it moved quickly to restore access while continuing to work with officials on broader availability. The company had said the original June 12 directive, which it said arrived at 5:21 p.m. ET, applied to any foreign national, including foreign-national employees inside the United States. Officials had justified the restriction by warning the models could be used by military intelligence users in China, Russia, or other countries of concern.

The policy shift comes after months of escalating concern around the model itself. Anthropic first made Mythos Preview available on April 7 through Project Glasswing to a limited group of tech companies, saying the system could identify and exploit zero-day vulnerabilities across major operating systems and browsers. The company later said more than 99% of the vulnerabilities it found had not yet been patched. In June, Anthropic paired Mythos 5 with Fable 5, pricing both at $10 per million input tokens and $50 per million output tokens, while describing Mythos 5 as an upgrade for a small group of cyberdefenders and infrastructure providers.

The new arrangement also raises a broader question about who gets to test powerful models after a security scare. Civil-liberties and free-speech advocates have criticized the government’s role in choosing customers, and OpenAI chief executive Sam Altman has said he did not like the government picking customers. Anthropic’s partial return for Mythos 5 suggests a playbook taking shape in Washington: restrict first, then reopen only to a vetted circle of firms, agencies, and infrastructure operators.