The court voided the risk label on Anthropic and enjoined the measures, but rejected the ultra vires claim

On August 27, Judge Rita F. Lin of the Northern District of California decided Anthropic PBC v. U.S. Department of War on the merits. The opinion runs to 59 pages and the operative order to four. Both are worth reading over the coverage, because the coverage says Anthropic won and the order says both sides did, in different places.

What the government put in the record

The challenged actions date to 27 February and 3 March 2026: the President and Secretary of War Hegseth designated Anthropic a supply chain risk to national security, ordered every federal agency to stop using its products permanently, and barred all defense contractors from doing any business with the company, including business unrelated to the military.

The court's description of the justification is the part to sit with. The administrative record is, in the opinion's word, slim: a single four-page memorandum, which postdates two of the three actions it is meant to justify, carries the entire government rationale.

The substance of that memorandum did not survive. Its risk assessment rested on Anthropic having backdoor access to its technology once deployed in a national security system. The court records that Anthropic undisputedly has no such access, that the government has backed away from the claim, and that defendants concede the technology is no riskier to national security than any other black-box model.

What was left was trust, and trust was the problem

With the technical case gone, one factor unique to Anthropic remained: trust. The government's position was that because of what it called Anthropic's increasingly hostile manner through the press, and the company's criticism of the Department's views on AI use, it could not trust Anthropic to ensure the integrity of its models.

The court did not weigh that as a close question. It found the actions were taken from a desire to make a public example of Anthropic for its arrogance in criticizing the government, and not from any articulable basis for believing the company would sabotage its own model.

Three facts in the record did that work, and none of them came from Anthropic.

• Days before the designation, Secretary Hegseth had proposed applying the Defense Production Act to Anthropic, which treats a company as essential to national security rather than as a threat to it.
• Immediately after the designation, the Department of War went on pursuing a contract with Anthropic, in its own words very close here.
• The government is still discussing collaboration with Anthropic on its newer model, Mythos, across an array of sensitive contexts.

The order is split, and the split matters

Both sides' motions were granted in part and denied in part. Anthropic won the First Amendment count and the Fifth Amendment due process count, and won its Administrative Procedure Act challenge to the Hegseth Directive and the designation against Secretary Hegseth and the Department of War.

It also lost, in three places. The ultra vires count went to the government outright. On the Section 558 count Anthropic prevailed against nine agencies and lost against five more, among them Health and Human Services, Commerce, Veterans Affairs, the SEC and NASA. And against what the order calls the Non-Participating Defendants, which include the Executive Office of the President, the Federal Reserve Board of Governors and the Social Security Administration, the government won on every claim.

The relief is nonetheless broad. The challenged actions are permanently enjoined; the defendants must rescind the guidance and directives that carried them out; the portion of the Hegseth Directive barring contractors from any commercial activity with Anthropic is vacated as arbitrary, capricious and an abuse of discretion; and the implementing actions of nine named agencies are vacated as orders imposed without delegated jurisdiction. That vacatur does not reach the actions the Federal Housing Finance Agency took in its capacity as conservator.

One paragraph keeps the ruling honest about its own limits. Nothing in it bars any lawful action that was available on February 27 2026, it does not require the Department of War to use Anthropic's products, and it does not stop the Department from moving to another AI provider, so long as those actions comply with the applicable regulations, statutes and constitutional provisions. The court retained jurisdiction to enforce.