Anthropic’s attorneys told a federal appeals court that they cannot “manipulate” their Claude artificial-intelligence system once it is deployed inside classified Pentagon military networks, framing the argument as part of a bid to defeat the Trump administration’s efforts to characterize the company as a security risk. The company made the assertion in a 96-page submission filed in Washington D.C., where the court is considering its challenge to Pentagon actions that, according to Anthropic, carry a stigma tied to national-security safeguards.
The lawsuit stems from a contract dispute in which the Pentagon and Anthropic disagreed about how AI technology could be used in military systems, including scenarios described in the court papers involving fully autonomous weapons and potential surveillance. Anthropic says the government is using a designation intended to protect against sabotage of national security systems by foreign adversaries to retaliate against the company rather than address an actual security concern.
The filing also points to timing and procedure before the court. Earlier this month, the appeals court denied Anthropic an order that would have paused the Pentagon’s actions while the panel continued to collect evidence in the case. In the new submission, Anthropic said it is attempting to address questions from the court ahead of oral arguments scheduled for May 19, when the Trump administration will have another chance to respond.
Anthropic’s temporary setback in the Washington case follows a separate legal fight in federal court in San Francisco that addressed similar issues. In that earlier matter, Anthropic prevailed, and the court said the decision prompted the Trump administration to remove “stigmatizing labels” from the company, according to court filings.
Even so, Anthropic said the parallel case in Washington remains unresolved, leaving what it described as a lingering cloud over the company. The dispute has also affected Anthropic’s business standing with the government: after the Pentagon canceled a $200 million contract with Anthropic in the wake of their disagreement, OpenAI moved in to provide its technology to the U.S. military.
For Anthropic, the appeals case is now focused on persuading the court that its Claude tool cannot be altered once it is deployed in classified environments—and that the Pentagon’s labeling effort is not a lawful response to those capabilities. For the administration, the next round will come after the filing, when it submits its reply before the May 19 hearing and the panel decides whether the Pentagon’s actions should be upheld while the case proceeds.