Anthropic’s Pentagon deal draws split rulings

A federal appeals court in Washington let the Pentagon keep Anthropic on a national-security blacklist on April 8, even though a federal judge in California had blocked much of the same crackdown on March 26. The result is a split-screen legal mess: one court says the government likely overreached, another says the blacklist can stay in place while the fight continues. (media.cadc.uscourts.gov) (cases.justia.com) This started after Anthropic refused Pentagon contract language that would allow “all lawful use” of its Claude model. Anthropic had pushed to keep two limits in place: no domestic surveillance of Americans and no fully autonomous weapons. (cbsnews.com) (breakingdefense.com) President Donald Trump then ordered federal agencies on February 27, 2026 to stop using Anthropic’s technology, and Defense Secretary Pete Hegseth posted that no military contractor, supplier, or partner could do commercial activity with Anthropic. On March 3, the Pentagon formalized that move by labeling Anthropic a “supply-chain risk.” (breakingdefense.com) (cases.justia.com) A supply-chain risk label is procurement law’s version of telling every buyer in a giant company not to touch one vendor. It does not just block a direct Pentagon contract; it can also scare off outside contractors that need the Pentagon’s approval to keep using that vendor’s tools. (politico.com) (cbsnews.com) Judge Rita Lin in San Francisco said on March 26 that Anthropic was likely to succeed on the merits and wrote that the record suggested the government’s stated reasons were “pretextual.” Her order blocked enforcement of the presidential directive, the Hegseth directive, and the March 3 designation under the Defense Department statute. (cases.justia.com) (breakingdefense.com) But lawyers warned that win was incomplete because Anthropic had also filed a separate case in the District of Columbia over a different procurement statute. Politico reported that three contractors had already cut ties or been told to do so, and three deals worth more than $180 million had fallen apart while the label was hanging over the company. (politico.com) That second front is where Anthropic lost, for now. The District of Columbia Circuit said Anthropic had not met the “stringent requirements” for an emergency stay, but it also fast-tracked the case with briefs due April 22 and May 6 and oral argument set for May 19, 2026. (media.cadc.uscourts.gov) (cnbc.com) The appeals panel framed the balance in stark terms: financial harm to one company on one side, and “how, and through whom,” the Pentagon secures artificial intelligence during an active military conflict on the other. That tells every artificial intelligence vendor what argument will dominate these cases once the words “national security” are on the table. (law.com) (media.cadc.uscourts.gov) The strange part is that the courts are not yet deciding whether Anthropic’s safety limits were wise. They are fighting over procedure, jurisdiction, and which statute controls which government action, which is why one order can freeze a ban while another lets a blacklist keep biting. (media.cadc.uscourts.gov) (cases.justia.com) That legal split is already changing the market. Defense News reported that smaller artificial intelligence firms began getting calls from generals and investors after Anthropic’s ouster, because procurement officers do not wait around when a core supplier is tied up in court. (defensenews.com) So the case is no longer just about Anthropic or Claude. It is becoming a test of whether a commercial artificial intelligence company can sell to Washington, draw bright red lines around military use, and still survive a procurement system built to treat uncertainty itself as a threat. (wired.com) (politico.com)