Anthropic loses appeal on risk designation

A federal appeals court in Washington refused to pause the Pentagon’s “supply chain risk” label on Anthropic on April 8, so the company stays shut out of new Defense Department contracts and Pentagon systems while the case continues. The judges did not end the case, but they said Anthropic had not met the high bar for an emergency stay and would instead get a fast-tracked appeal on the merits. A “supply chain risk” label is the government’s way of saying a vendor is too risky to sit inside sensitive systems, like a building manager deciding one contractor can’t have keys to the server room. Once that label lands, procurement officers do not wait for a long trial before changing who gets access. This fight is not mainly about whether Claude is a strong chatbot. It is about whether the government can treat Anthropic itself as a risky supplier after a clash over military access to the model. The dispute started after Anthropic refused to give the United States government unrestricted military access to Claude, according to multiple reports on the case. The Trump administration then moved in February to stop federal agents from using Claude and later formalized the Pentagon risk designation in early March. That created two different legal tracks at once. A federal judge in San Francisco blocked a broader governmentwide ban on Claude in late March, but the Washington appeals court still left the Pentagon’s separate supplier-risk label in place. So Anthropic is in the strange position of being allowed to keep fighting the wider restrictions while still being frozen out of new Defense Department business right now. That is why this ruling stings even though it is only an interim loss. The judges signaled the tradeoff in blunt terms. Law.com reported the panel weighed “financial harm to a single private company” against court interference in how the Pentagon secures artificial intelligence during an active military conflict. That line explains why this case reaches beyond one startup. If the government can cut off a model maker by labeling the company part of the risk, access to huge institutional customers can turn on national-security discretion faster than on product benchmarks or market share. Anthropic still gets its faster appeal, and the panel acknowledged the company could suffer “irreparable harm” while the case plays out. But as of April 10, the Pentagon’s label is still live, and that means the doors that matter most in Washington remain closed for now.