Anthropic told an appeals court it cannot manipulate Claude once the model is deployed within classified military networks. The company is fighting a Trump administration effort to label it a supply chain risk. This legal battle determines whether AI labs maintain control over weights after deployment in secure government environments.