Federal Judge Rules Trump Administration’s Anthropic Blacklist Illegal

The dispute between Anthropic and the Trump administration has become a test of how far the US government can go when an AI company rejects certain military and surveillance uses of its products. Judge Rita Lin of the US District Court for the Northern District of California ruled that the administration acted unlawfully when it designated Anthropic a national-security supply-chain risk and directed the federal government to stop using the company’s technology.

What the ruling says

The case followed Anthropic’s refusal to remove restrictions on the use of its Claude models for lethal autonomous warfare and mass surveillance of Americans. The administration responded by ordering federal agencies to permanently stop using Anthropic products. It also barred defense contractors from doing business with the company, including work unrelated to the military.

In granting key portions of Anthropic’s summary-judgment motion, Lin wrote that the undisputed record showed unlawful retaliation in violation of the First Amendment. Her order vacated the government actions and required the administration to rescind the directives that the court found illegal.

The ruling rests on several points:

• The Defense Department is free to select the AI vendor it wants, but that discretion does not authorize an across-the-board punishment of a company for criticizing government policy.
• A general invocation of national security is not a blank check for retaliation against government critics.
• The relevant supply-chain-risk law addresses covert sabotage, malicious functions, or other forms of subversion of a covered system—not a company’s openly stated safety policies or contract terms.
• The government’s justification was thin. It had moved away from an earlier theory that Anthropic could retain backdoor access after deployment, and it conceded that Anthropic lacked such access and was no riskier than other black-box AI models.

Lin also held that the administration’s actions were arbitrary and capricious under the Administrative Procedure Act. In her reading, Anthropic did not meet the statutory definition of a supply-chain risk merely because it publicly opposed particular uses of its systems.

Broader implications

The decision links AI safety commitments with constitutional protections and public procurement rules. Anthropic argued that it has the right to express views about the limits of its services, both publicly and in discussions with the government, and to impose safeguards based on its assessment of AI risks. The court’s reasoning suggests that the government may decline to buy a company’s products, but cannot automatically recast the company as an adversary because it will not accept certain contract conditions.

That distinction matters well beyond Anthropic. Technology companies that sell to the federal government may now have a stronger basis to challenge attempts to bypass ordinary procurement procedures and impose sweeping commercial bans under a national-security label. The Computer & Communications Industry Association welcomed the outcome, saying the case matters to any company doing business with the US government.

The legal fight is not necessarily over. The administration previously obtained an early ruling from the US Court of Appeals for the District of Columbia Circuit denying Anthropic’s emergency request for a stay. That court later heard oral arguments and is still reviewing the broader dispute. The administration could also appeal Lin’s latest order.

The government’s practical use of Anthropic’s tools has not completely ended, according to the source report. Agencies reportedly continued using them while the White House and Anthropic gradually repaired relations ahead of the company’s planned IPO. The eventual outcome will help define how US agencies classify AI supply-chain risks—and whether companies can maintain safety restrictions on military and surveillance applications without risking government retaliation.

Source: Ars Technica AI