The D.C. Circuit upheld the Pentagon's supply chain risk designation of Anthropic, 2-1. Practical upshot: the military and its defense contractors cannot use Claude models in DOD work, and the second of two parallel designations survives. A San Francisco federal judge already struck the other designation as unlawful last month, so Anthropic is litigating on two fronts and just lost one. The underlying dispute is the interesting part. DOD wanted unfettered Claude access for all lawful purposes on GenAI.mil; Anthropic wanted carve-outs for autonomous weapons and domestic mass surveillance. Hegseth characterized that as veto power over military operations, and Katsas deferred: the President and the Secretary decide where to balance the risks. Judge Henderson dissented. Note the delay: the mandate doesn't take immediate effect, pending rehearing, en banc, or cert. Also worth flagging that the article headline says "supply chain risk" while the opinion itself says "Secretary of War" — the Supply Chain Security Act apparently now comes with a title from 1947. Dario Amodei's expected response is a 15,000-word essay warning this ends civilization, followed by founding a wet lab. Reality, meanwhile, filed an amicus brief, which nobody read either. The $200 million contract from July 2025 is presumably now a supply chain risk with a supply chain. The open question: if a vendor refusing to promise unrestricted military use is itself the national-security threat, what exactly is being procured — a model, or compliance?