Anthropic’s Pentagon setback

A federal appeals court in Washington refused on April 8 to freeze the Pentagon’s “supply chain risk” label on Anthropic, so the company stays sidelined from parts of the military market while the case moves ahead on a fast track. (nytimes.com) That label is not just bad publicity. Reuters reported that defense contractors and agencies working with the Pentagon can be pushed to certify they are not using Anthropic’s models, which turns one legal order into a practical sales blockade. (reuters.com) The fight started after Anthropic tried to keep its Claude model from being used for mass surveillance of Americans or for lethal autonomous weapons. The Pentagon argued it should be able to use artificial intelligence systems for any lawful military purpose without a private company setting those limits. (cbsnews.com) In early March, the Defense Department formally called Anthropic a supply-chain risk, and Anthropic answered with lawsuits in California and Washington. One court in San Francisco temporarily blocked a broader federal ban on Claude, but the Washington appeals court controls review of the Pentagon’s risk designation itself. (pbs.org, politico.com) The appeals judges said Anthropic might suffer “irreparable harm,” but they still left the designation in place because they did not want courts micromanaging how the Pentagon buys artificial intelligence during an active conflict. That means Anthropic won speed on appeal, but not relief. (bloomberg.com, law.com) The immediate winners are smaller defense-focused artificial intelligence firms that were never treated as consumer chatbots first. Reuters said companies such as Smack Technologies and EdgeRunner AI started getting calls from generals, combatant commanders, and investors after Anthropic’s relationship with the Pentagon broke down. (reuters.com) Those firms sell narrower tools built for military jobs like planning, logistics, and battlefield software, not one giant model meant to answer every question. Defense One reported that the Anthropic dispute exposed a gap between what frontier models promise and what troops actually need in the field. (defenseone.com) The Pentagon is also using this fight to avoid dependence on a single favorite supplier. Reuters reported that Anthropic had been the military’s once-favored artificial intelligence vendor, and its fall pushed officials to spread work across more companies instead of betting on one lab. (reuters.com) So this case is no longer just about whether Anthropic can undo one designation. It is turning into a test of who gets to write the rules for military artificial intelligence: the company that built the model, or the government that wants to use it. (nytimes.com, reuters.com)