SAN FRANCISCO-A federal judge ruled late Thursday that the Trump administration violated Anthropic's First Amendment rights by blacklisting the artificial-intelligence company as a supply-chain risk.

U.S. District Judge Rita F. Lin of the Northern District of California mostly sided with Anthropic in her ruling on the company's March lawsuit against the Trump administration, saying the government had erred in its haste to give Anthropic that designation.

The government didn't provide evidence that the company was an actual national security risk, Lin wrote in her decision, but rather acted in "unlawful retaliation" to Anthropic's public pushback against the Pentagon's use of AI.

"Defendants' contemporaneous words and deeds confirm that the challenged actions were based on a desire to make a public example out of Anthropic for its 'arrogance' in criticizing the government," Lin wrote.

She directed the government to rescind all guidance, directives, communications, or instructions to blacklist and label Anthropic a supply-chain risk.

"We welcome the court's ruling that this supply-chain risk designation was unlawful," an Anthropic spokeswoman said. "We remain focused on working productively with the government to harness AI for our national security so all Americans benefit from this technology."

The Defense Department didn't immediately respond to a request for comment.

The ruling comes at a critical time for Anthropic. The company has vaulted ahead in the AI race by many measures and is barreling toward an initial public offering in which it is aiming for a valuation of about $2 trillion, yet its clashes with the Trump administration have remained a risk. Government lawyers said the Defense Department would be entirely moved off Anthropic models by Sept. 30. Government correspondence filed as part of the case shows competitors including OpenAI are moving to take Anthropic's place.

Anthropic in June struck an agreement with the administration to release its more powerful models, Fable and Mythos, after a shutdown caused by security concerns. U.S. government agencies outside the Defense Department are using those and other tools from Anthropic.

Lin said the continued use of Anthropic models by the government was in conflict with the Pentagon's assertion that the company is a national-security risk.

Lin, who was appointed to the federal bench by President Joe Biden, granted Anthropic a preliminary injunction in March to stop parts of the supply-chain risk designation and other administration blacklistings.

She made clear during court hearings that she was unconvinced and even concerned by the government's legal arguments. At a July hearing, Lin called the government's position that it is legally permissible to retaliate against a contractor that publicly criticizes the administration "really troubling" and "quite extreme."

The administration argued that it hadn't infringed on Anthropic's free-speech protections and was within its rights to exercise its authority over a government vendor. The government said Anthropic is an untrustworthy vendor that may surreptitiously modify its models to inhibit military operations.

"The Department of War needs to trust that its AI vendors are going to be forthright and honest with the department," government attorney James Harlow said during court arguments.

Anthropic's Claude model had been widely used by the U.S. military, including in its January raid in Venezuela and the war with Iran. But after protracted and public negotiations to agree on a new contract to use the AI models in a military setting failed, Anthropic found itself in the Pentagon's crosshairs.

Earlier this year, Anthropic sought assurances from the Pentagon that its models wouldn't be used in fully autonomous weapons or for domestic surveillance. The Pentagon pushed back, wanting an agreement in which the military could use the AI in all legal applications, and considered Anthropic's demands too restrictive.

Anthropic said the government's actions had contributed to the loss of hundreds of millions of dollars in canceled, truncated or stalled contracts.

The Trump administration, which could appeal Lin's decision, is also awaiting a ruling from a Washington federal appeals court in a separate case on the issue. That court has been less receptive to Anthropic's arguments and in April denied the company's request for relief from aspects of the designation.