The news
In the Anthropic Pentagon ruling of August 27, 2026, U.S. District Judge Rita F. Lin found that the Department of War's supply chain risk label on Anthropic was unlawful retaliation for the AI developer's public criticism, and granted it summary judgment on most claims.
The case, filed on March 9 in the U.S. District Court for the Northern District of California, challenged actions taken on February 27 and March 3, 2026. According to Lin's 59-page order, President Donald Trump and Secretary of War Pete Hegseth designated Anthropic a supply chain risk, ordered federal agencies to permanently stop using its products and barred defense contractors from doing any business with the company, even work unrelated to the military.
The fight began over usage limits. Anthropic, led by CEO Dario Amodei, had insisted that the government agree not to use Claude for fully autonomous lethal weapons or mass surveillance of Americans, according to the court's March preliminary injunction order. Lin found that the measures violated the First Amendment and denied Anthropic the pre-deprivation process the Fifth Amendment requires, and that Hegseth's decision to designate Anthropic a supply chain risk violated 10 U.S.C. § 3252, the governing statute, and was arbitrary and capricious under the Administrative Procedure Act.
The judge wrote that the government's record was slim: a four-page memorandum supplied its whole rationale, and the government conceded Claude was no riskier to national security than any other black-box AI model. She noted the department kept pursuing a contract with Anthropic after acting against it, and wrote that national security was "not a blank check to punish and retaliate against government critics."
Lin ruled for the government on Anthropic's separation-of-powers claim and denied the government's request for a seven-day administrative stay of the permanent injunction, noting it had complied with her March 26 preliminary injunction for more than five months. According to analyses by the law firms Taft and Jones Walker, the relief vacated the Section 3252 designation and the bar on defense contractors doing business with Anthropic. The order did not force the Pentagon to use Claude; Lin wrote that the department remained free to choose its AI vendor and to lawfully end its contracts.
The ruling did not cover a second designation under the Federal Acquisition Supply Chain Security Act (FASCSA), 41 U.S.C. § 4713, which allows exclusions from agency contracts and subcontracts. Anthropic challenged it directly in the U.S. Court of Appeals for the D.C. Circuit, which denied a stay on April 8 and heard argument on May 19. Anthropic welcomed Lin's ruling, TechCrunch reported; Nextgov said the Pentagon had not returned a request for comment by publication time.
The numbers
- Length of summary judgment order
- 59 pages
- Dates of challenged actions
- February 27 and March 3, 2026
- Anthropic's 2025 Pentagon AI agreement (per March order)
- Up to $200 million over two years
- Revenue drop Anthropic said it faced if measures returned
- 50%-100% of defense contractor and other department-related client revenue
Why CEOs should care
For defense contractors, the ruling made permanent what Lin's March 26 preliminary injunction had already done: it lifted the broadest threat, that using Claude anywhere in the business, even for a customer service chatbot, could cost them Pentagon work. Procurement and legal teams that paused Anthropic tools after March should revisit those decisions but keep Pentagon work separate. Because the FASCSA exclusion remained in force while the D.C. Circuit deliberated, using Anthropic products to perform covered Department of War contracts still carried risk.
For CFOs and boards, the case shows how quickly an AI supplier can become a political target. According to unrebutted evidence cited in the order, reinstating the measures would cut Anthropic's defense contractor and other department-related client revenue by 50 to 100 percent and its 2026 revenue by multiple billions of dollars. Boards should ask whether critical workflows can switch models if a vendor is suddenly barred, and whether contracts address government-ordered exclusions.
For CISOs and general counsel, the court's findings are useful in vendor-risk reviews: Lin found Anthropic cannot access, alter or shut down models once deployed on government systems. They do not answer the question at the center of this dispute, which is whether a vendor's usage policies fit the customer's intended uses. Map those policies against your own use cases before signing.
The bigger picture
According to the court's March order, the supply chain risk label had never before been applied to a domestic company; it is aimed mainly at foreign intelligence services, terrorists and other hostile actors. Lin drew a line between the government declining to buy a product, which it may do, and punishing a supplier for criticizing it.
That line matters to every AI vendor with an acceptable-use policy. The case tests whether developers can restrict military uses of their technology without being shut out of the wider defense market.
What happened next
On September 25, the D.C. Circuit ruled 2-1 against Anthropic in the parallel case, denying its petitions for review and upholding the FASCSA exclusion. Judge Gregory Katsas, joined by Judge Neomi Rao, wrote that the department had ample support for concluding that continued use of Claude in its systems, by the department or its contractors, posed a covered national security risk. Judge Karen LeCraft Henderson dissented. The majority said it had no quarrel with the California court's reading of Section 3252.
Pentagon spokesman Sean Parnell said the decision validated the department's position, while Anthropic said it was considering all options, including further review, the Associated Press reported. The AP said contractors cannot use Anthropic products for Pentagon work, though Claude remains available elsewhere in government. Watch for any rehearing or Supreme Court request, and for whether the government appeals Lin's judgment to the Ninth Circuit.
