Appeals Court Upholds Pentagon Risk Label on Anthropic

A divided federal appeals court in Washington, D.C., upheld one of the Pentagon’s legal designations of Anthropic as a supply-chain risk, allowing the Defense Department to continue excluding the company’s Claude artificial intelligence models from its systems and barring contractors from using them for Defense Department work. The 2-1 ruling on September 25 rejected Anthropic’s challenge under the Federal Acquisition Supply Chain Security Act. The majority said the department had sufficient grounds under that law to conclude that Claude’s continued integration into government information systems could present a covered national-security risk. The judges pointed to restrictions Anthropic builds into the model and instances in which Claude had not performed tasks requested by government users. The law did not require the government to prove that Anthropic intended harm, the majority concluded. The dispute followed a breakdown in negotiations over the terms governing military use of Claude. Anthropic had sought to prevent use of its technology for lethal autonomous warfare and domestic surveillance, while the department sought permission to use the models for lawful purposes.