The legal posture is now weird enough to summarize. In early March, Defense Secretary Pete Hegseth designated Anthropic a “supply chain risk,” the bureaucratic equivalent of escorting a vendor off the loading dock with a security guard. The trigger was not a leaked model or a failed eval. The trigger was that Anthropic kept enforcing two policies it has carried since founding: no fully autonomous weapons that target humans without a human in the loop, and no mass surveillance of Americans. The Pentagon wanted Claude for “all lawful purposes.” Anthropic said the two lists overlap less than the Pentagon thinks. Hegseth picked up his phone.
A federal judge in California then ruled the designation violated Anthropic’s First Amendment and due-process rights and blocked it. A DC Circuit panel, on the same paperwork, declined to stay the designation while the case plays out, which split the courts at the trial level and sent the substantive question up on an expedited schedule. On May 19, that DC panel heard oral arguments. Judge Henderson, a George H. W. Bush appointee, said on the record she sees no evidence Anthropic actually poses a supply-chain risk. Judge Rao, a Trump appointee, asked what basis any court has for second-guessing the Secretary of Defense’s national-security judgment. Judge Katsas sat between them. A ruling is pending and could land before the end of summer.
Underneath the docket, the Pentagon is already moving on. Bloomberg confirmed in late May that 25 designated “power users” are putting OpenAI, Google, and xAI’s Grok through their paces inside GenAI.mil, which is the standalone evaluation platform DoD spun up specifically to route around Anthropic. The $200 million contract Anthropic signed last July is the thing actually at stake; Trump’s March executive order told federal agencies to phase out Anthropic over six months, which puts the cliff in early September.
The structural question the courts are now being asked, in plainer language than either filing puts it, is whether a frontier-model lab is allowed to refuse a government customer over the use case and not get banned for the refusal. Anthropic’s answer is yes. The Pentagon’s answer is no. Everybody else in the industry is watching this one quietly because the precedent applies to all of them.