The US government has lifted its export block on Anthropic's Claude Mythos 5, allowing the model to be released to over 100 US institutions including major companies and government agencies. Commerce Secretary Howard Lutnick issued a letter stating that a license is no longer required for exporting Mythos 5 to entities listed in Annex A, reversing a two-week standoff. The move follows concerns over Chinese access, after SK Telecom, a South Korean carrier, was added to the Project Glasswing vulnerability-hunting program and then lost access. Senior Anthropic staff had flown to Washington to meet officials during the dispute. Meanwhile, the consumer version Fable 5 remains offline, though sources indicate its release is advancing. The episode signals a new AI gatekeeping system, where frontier models are subject to federal review. OpenAI also limited its GPT-5.6 tier Sol to about 20 government-approved partners. Dozens of cybersecurity leaders, including from Nvidia, Adobe, and Zoom, had urged the administration to drop the controls. Allies abroad have grown frustrated with dependency on US access decisions.

Key facts

• US govt lifts export block on Anthropic's Claude Mythos 5 for 100+ US institutions
• Commerce Secretary Lutnick says license no longer required for Annex A entities
• Consumer version Fable 5 remains offline; sources say release is advancing
• Project Glasswing vulnerability program included SK Telecom, later lost access
• OpenAI limits GPT-5.6 Sol to ~20 gov partners; weaker versions public

KeyAudit data perspective