Anthropic loses stay

A federal appeals court in Washington left Anthropic under a Pentagon “supply chain risk” label on April 8, refusing the company’s request to freeze the designation while the case continues. The same order also sped up the next phase, telling both sides to brief the merits on an unusually fast schedule instead of letting the fight drag for months. (politico.com) That sounds procedural, but a stay is basically a pause button. Anthropic asked the court to stop the government from enforcing the label now, and the court said no, so the label keeps operating while judges move quickly toward a fuller review. (cnbc.com) The label came from the Department of Defense in early March after a dispute over guardrails on military uses of Claude, Anthropic’s flagship language model. Defense Secretary Pete Hegseth said Anthropic’s conduct created a national security problem and barred the Pentagon and its contractors from using the company’s artificial intelligence products on Defense Department work. (usnews.com) A supply chain risk designation is not a criminal charge and it is not the same thing as banning a consumer app from the internet. It is closer to a government procurement blacklist, because it tells defense agencies and outside contractors that a vendor is too risky to sit inside sensitive systems or workflows. (computerworld.com) Anthropic says the Pentagon is stretching that power far beyond what the law allows. In public statements last month, the company argued that the designation cannot legally cut off unrelated commercial work and said it would challenge any attempt to turn a defense contracting tool into a broader punishment for the company. (anthropic.com) The company also says the fight started because it tried to keep some military uses inside its own safety rules rather than hand over a blank check. Chief executive Dario Amodei wrote in March that Anthropic had not objected to military work in general, but did object to a narrow set of uses it believed could undermine democratic values or exceed what current systems can safely do. (anthropic.com) That is why this case is bigger than one court motion. It is testing who gets the final say when an artificial intelligence company sells tools to the government but still wants limits on how those tools are deployed in war planning, targeting, or other high-stakes operations. (nytimes.com) The legal picture is also messy because Anthropic won a separate ruling in San Francisco in late March. A federal judge there issued a preliminary injunction blocking enforcement of a different Trump administration ban on Claude, which left contractors and customers staring at two federal courts sending different signals at almost the same time. (cnbc.com) Wednesday’s order does not decide whether the Pentagon was right on the facts or the law. It decides that Anthropic did not clear the high bar for emergency relief, and it moves the case onto a faster track where the court can answer the bigger question sooner. (politico.com) For now, the practical effect is simple: defense contractors that want Pentagon business still have to treat Anthropic as flagged while the appeal runs, even as Anthropic keeps arguing that the government’s reading of its own power is unlawful. A case that started as a fight over one company’s model policy is turning into an early test of how much leverage Washington can use over the private artificial intelligence supply chain. (computerworld.com)