Anthropic's red line survived the Pentagon

Anthropic refused to let Claude be used for mass domestic surveillance or fully autonomous weapons. The Pentagon answered by designating the company a supply-chain risk. On August 27, U.S. District Judge Rita Lin ruled that designation illegal and baseless. It is a major win for Anthropic, but not the end of the dispute.

What Anthropic refused

This was not a refusal to work with the military. Anthropic supported lawful foreign intelligence and counterintelligence missions, but argued that current frontier models are not reliable enough to select and engage targets without human oversight. It also said AI-powered mass surveillance could assemble an intimate picture of Americans' lives at enormous scale.

How a contract fight became a blacklist

The Pentagon wanted contract language allowing Claude for "all lawful purposes." Anthropic said that could permit the two uses it had excluded. After negotiations collapsed, President Trump directed federal agencies to stop using Anthropic's technology, and Defense Secretary Pete Hegseth designated the company a supply-chain risk under 10 U.S.C. § 3252. Contractors were told to identify, remove, and certify non-use of Anthropic products on covered work.

What the judge decided

Lin granted Anthropic summary judgment in the California case. She found that the government's own words and actions showed a desire to make a public example of the company for criticizing its AI policy, not an articulable basis to believe Anthropic would sabotage its model. The designation was unlawful First Amendment retaliation, and the process also violated Fifth Amendment due process.

The ruling does not require the Pentagon to buy Claude. The government can choose another vendor; it simply cannot turn national security into a blank check for punishing a company that criticizes its contracting position.

The catch

Anthropic also filed a separate challenge in Washington under the Federal Acquisition Supply Chain Security Act. The D.C. Circuit has not paused that designation, so the California judgment does not automatically restore a clean path into every federal contract. Contractors may still face certification and compliance questions while the second case proceeds.

Why it matters

Safety policies are no longer just product rules. They are procurement terms. Anthropic says a vendor should not knowingly provide an unreliable system for autonomous lethal decisions or population-scale surveillance. The Pentagon says a government customer should not accept unilateral limits on lawful national-security work.

The next signals are the government's appeal, the D.C. Circuit's ruling, and the contract language that defines human oversight. The court did not settle how much control a model maker should have over military use. It did settle, for now, that retaliation cannot be disguised as a supply-chain finding.

The point

The ruling doesn't give Anthropic the right to sell Claude to the Pentagon. It protects something narrower: the government can't punish an AI company for its safety limits and call that punishment a national-security procurement decision. The case that is still open will decide whether that protection holds outside California.