A California court determined that the Pentagon’s labeling of Anthropic as a supply-chain risk was an unlawful, “arbitrary,” and “capricious” act designed to discipline the AI company.
Anthropic was never a legitimate national security risk.
That is the conclusion of the ruling handed down Thursday by the U.S. District Court for the Northern District of California. Judge Rita Lin held that the Trump-era administration violated the First Amendment and unlawfully retaliated against Anthropic for engaging in “constitutionally protected expressive activities” when it designated the firm as a supply-chain risk. Lin also determined that the government’s actions breach the Fifth Amendment’s Due Process Clause by “depriv[ing] Anthropic of constitutionally protected liberty interests without adequate pre-deprivation notice or opportunity to be heard.”
The clash between the AI company and the government began in February, when Defense Secretary Pete Hegseth and Anthropic’s chief executive Dario Amodei exchanged public remarks about how the Pentagon was using Anthropic’s models. Amodei refused to yield to Hegseth’s demand that Anthropic’s model be “free from usage-policy constraints that could limit lawful military applications.” Anthropic insisted that its contract with the Pentagon include limits on deploying its technology for fully autonomous weapons and mass surveillance. Hegseth then threatened to nationalize the company and label it a supply chain risk.
When Anthropic remained noncompliant, President Donald Trump ordered every federal agency to cease using the company’s technology. Hegseth instructed Defense Department contractors, suppliers, and partners to halt commercial dealings with Anthropic, branding it as a supply chain risk.
That directive, Lin found, was “arbitrary, capricious, and in excess of statutory jurisdiction, authority, or limitations” under the Administrative Procedure Act. Lin noted that Hegseth acted “without observing the procedure required by law.”
The initial order from Trump extending Anthropic’s designation as a supply-chain risk to other federal agencies—such as the Treasury Department and the Office of Management and Budget—was also struck down. Lin ruled that the president had imposed these conditions “without the jurisdiction delegated to those agencies and without statutory authorization.”
Lin issued a permanent injunction barring the Defense Department from “implementing, enforcing, or giving effect to” Hegseth’s directive that cast Anthropic as a supply-chain risk. She demanded that the department rescind its prior “guidance, directives, communications, or instructions” related to the designation and take “all necessary actions” to ensure the designation was not put into effect.
In a twist of irony, the Pentagon’s own language undermined its case. In her 59-page opinion, Lin notes that “it is undisputed Anthropic was granted high levels of security clearance and was consistently praised,” and there was “no discussion of Anthropic’s untrustworthiness or any potential supply-chain risk anywhere in the record” until the dispute with the Defense Department became public.
Things didn’t help the government when the Defense Department could not answer whether “terminating a contract for the undisputed sole purpose of ‘stop[ping] public criticism of the President’ would violate the First Amendment.” Yet federal courts have already addressed that question. “Criticism of government sits at the very core of the constitutionally protected domain of free discussion,” Lin emphasizes.
It isn’t a flawless win for Anthropic, though. There remains a pathway for the Trump administration to sever the company’s technology from federal use. Lin clarifies in her order that the government is not blocked from “taking any lawful action that would have been available” prior to the designation.
Anthropic’s refusal to permit mass surveillance with its technology did not, on its own, establish the company as a national security threat. Government skeptics recognized that immediately, and now, to a welcome extent, the courts concur as well.
