Anthropic Blacklist Standoff
A federal appeals court declined, for now, to block the Pentagon’s national‑security blacklisting of Anthropic, leaving uncertainty about military access to the company’s Claude models. The ruling adds a supply‑chain risk layer to any programme that would rely on a commercial foundation model for defence use (marketscreener.com). Journalistic coverage notes conflicting rulings have created limbo around whether and how the US military can depend on Anthropic, making vendor-selection a procurement and legal question as much as a technical one (wired.com).
A federal appeals court in Washington left the Pentagon’s blacklist of Anthropic in place on April 8, which means defense contractors can still be told not to use Claude while the lawsuit keeps going. The judges did not decide the whole case; they only said Anthropic had not cleared the high bar for emergency relief. (apnews.com) The practical effect is simple: if you build software for the Department of Defense, a label on one company can suddenly make your favorite model off-limits. Politico reported that the Pentagon’s “supply chain risk” designation blocks contractors working on Department of Defense contracts from using Anthropic’s models. (politico.com) That phrase, “supply chain risk,” usually sounds like bad bolts or suspect chips. Here it is being used on a software supplier, which turns a chatbot into something the government treats more like a critical component in a weapons system. (mayerbrown.com) This fight did not start with a random audit. It grew out of a July 2025 Pentagon contract under which Claude became the first frontier artificial intelligence model approved for use on classified government networks, and then the relationship broke down during renegotiations. (mayerbrown.com) The break was about limits on use. Reporting from The Associated Press says Anthropic tried to stop Claude from being used for fully autonomous weapons and for mass surveillance of Americans, while Pentagon officials argued they could not let a vendor place those restrictions on a military capability. (apnews.com) In early March, the Defense Department formally designated Anthropic a supply chain risk, and CNBC reported that contractors were then required to certify they were not using Anthropic’s models in Pentagon work. That turned a policy dispute into an immediate procurement problem for every company in the defense stack. (cnbc.com) Then the courts split. A federal judge in San Francisco late in March granted Anthropic a preliminary injunction against a broader government ban on Claude, but the Washington appeals court this week refused to pause the Pentagon’s separate risk label. (cnbc.com) That is why this story is so messy for buyers inside the military. Bloomberg reported that the broader ban remains blocked for now, even as the supply-chain-risk declaration survives, so one court order says “not so fast” while another says “carry on.” (bloomberg.com) The Pentagon has used this kind of designation before, but not like this. Politico said Anthropic is the first American company to be publicly labeled a supply chain risk, which gives the case weight far beyond one vendor and one model family. (politico.com) That changes how defense buyers will think about artificial intelligence contracts. A model can be accurate, fast, and already integrated into secure systems, but if its legal status can flip in one week, the safer choice may be the vendor with fewer courtroom problems rather than the better benchmark score. (axios.com) Anthropic can still fight the case on the merits, and the San Francisco order still gives it some cover outside this narrow appeals ruling. But after April 8, any program manager planning around Claude has to treat court calendars and procurement rules as part of the software stack. (usnews.com)
