A California federal judge has blocked the Trump administration's punitive action against Anthropic, the developer of the Claude AI model, following a months-long legal battle with the Department of Defense over the company's restrictions on using its technology in warfare.

Judge Rita Lin ruled on Thursday that the Defense Department's designation of Anthropic as a supply chain risk was unlawful retaliation, a label typically reserved for enemies of the United States. The judge criticized the administration in her ruling, stating that "an empty invocation of national security does not amount to an unlimited license to punish and retaliate against government critics."

The conflict began earlier this year when the startup refused to sign an open-ended contract with the Pentagon. Anthropic's CEO Dario Amodei has maintained two firm "red lines": prohibiting the use of its technology in lethal autonomous weapons and in domestic mass surveillance. This stance drew the ire of Defense Secretary Pete Hegseth and President Donald Trump.

In late February, the Trump administration issued the supply chain risk determination, ordering all federal agencies to cease using Anthropic's products. That ban effectively prohibited all defense contractors from doing business with the AI lab, though it included a transition period allowing the Pentagon to continue using the technology during that time.

While Thursday's ruling is a victory for Anthropic, the company still faces a separate case before the Washington Court of Appeals. In April, that court denied Anthropic's request to pause the risk determination. The U.S. government may also appeal the California court's ruling.

The decision means federal agencies other than the Defense Department can continue using Anthropic's products. Several agencies have been testing the company's powerful Mythos large language model for analyzing cybersecurity and national security risks. In recent months, relations between the leading AI lab and some parts of the Trump administration have appeared to thaw, with close collaboration on the Mythos model's release. The model was rolled out cautiously over concerns its cybersecurity capabilities could be misused by bad actors.

In June, the government briefly blocked another Anthropic model, Fable, over security concerns before reaching an agreement with the company to allow its release. Anthropic welcomed the court's decision and said the company remains "committed to productive cooperation with the government to use AI to safeguard national security and benefit all Americans with this technology."