IC Technology & Surveillance — 2026-06-28

US Government Partially Reverses Mythos 5 AI Export Controls and Authorizes Access for Over 100 Approved Organizations After NSA Testing

BLUFSelective reversal on Mythos 5 cements government pre-clearance of frontier AI as a standing gate, making Fable 5 export relief unlikely by June 2027.

Commerce Secretary Howard Lutnick sent a letter Friday to Anthropic co-founder Tom Brown authorizing roughly 100 approved companies, federal agencies, and critical infrastructure operators to access Claude Mythos 5 without an export license, including for their foreign national employees 12. The letter, which both CNBC and Semafor reviewed, contains no provision for Fable 5, which remains restricted 12. Lutnick cited "significant progress" in two weeks of daily talks and Anthropic's commitment to work with the government on future release protocols 1. The approved organizations participate in Anthropic's Project Glasswing cybersecurity initiative; NSA and Cyber Command chief Gen. Joshua Rudd stated that during government testing, Mythos 5 identified critical vulnerabilities in classified government systems within hours 3.

Analysis
Friday's authorization establishes government pre-clearance of frontier AI as a durable regulatory feature, one in which Fable 5 is unlikely to receive export license relief by June 30, 2027. Moderate confidence rests on the administration's pattern of incremental concessions, the letter's deliberate silence on Fable 5, and Lutnick's framing of continued protocol development rather than a return to prior open access. NSA's public claim that Mythos 5 identified classified-system vulnerabilities within hours entrenches the security rationale for differential model treatment. Allied exclusion signals downstream friction in AI-sharing arrangements where Washington's approval authority has no treaty basis.

4 sources