Federal judge voids Pentagon blacklist of Anthropic, calls it unconstitutional retaliation
The AMW Read
A court reverses a national-security blacklist on a case-study frontier lab over weapons/surveillance use-case restrictions, setting a government-AI-procurement precedent beyond this one dispute.
Federal judge voids Pentagon blacklist of Anthropic, calls it unconstitutional retaliation
US district judge Rita Lin ruled Thursday that the Pentagon's February 27 designation of Anthropic as a "supply-chain risk" was illegal, vacating the label and a related ban on military contractors doing business with the company. Her 59-page ruling also struck down sanctions from nine federal agencies, including Treasury, State, and Homeland Security. The fight traces to a stalled $200 million Pentagon deal for Claude models: Anthropic wanted limits on lethal autonomous weapons and mass surveillance uses, while then-Defense Secretary Pete Hegseth demanded "all lawful use." Talks broke down after reports Claude was used in the US operation to capture Venezuelan president Nicolas Maduro. Lin noted Washington is simultaneously discussing using Anthropic's newer Mythos model in sensitive contexts, calling that inconsistent with sabotage claims. The Pentagon is expected to appeal; a separate Anthropic suit at the DC Circuit continues.
The ruling is among the first courtroom tests of whether Washington can use national-security procurement tools to punish a lab for refusing specific military use cases rather than for an actual security failure. It affirms, for now, that a lab can hold use-case restrictions without losing federal-contract eligibility. Per the AI Market Watch index, Anthropic-linked coverage in our pipeline nearly doubled to 285 items over the past 90 days versus 206 prior (coverage of pipeline-ingested sources, not a census), reflecting how contested this account has become across funding, compute, and legal fronts.
For builders and investors, the ruling trims near-term regulatory risk to Anthropic's government revenue just as it reportedly eyes an October IPO, and sits beside its separate $45 billion Nscale compute commitment -- legal exposure over use-case terms is now as material as compute access when pricing frontier-lab government risk.
