On August 27, 2026, Judge Rita F. Lin of the Northern District of California ruled largely in favor of Anthropic PBC in its suit against the Department of War and other federal agencies over the government’s decision to bar all federal use of Anthropic’s AI products and prohibit defense contractors from doing business with the company. The case, Anthropic PBC v. U.S. Department of War (No. 26-cv-01996-RFL), offers a rare judicial roadmap for how far the government can go when it decides a contractor is no longer “trustworthy”—and where that authority runs out.
The background
In February 2026, after Anthropic refused to drop contractual restrictions on using its Claude models for mass surveillance of Americans and lethal autonomous warfare, Secretary of War Pete Hegseth and President Trump publicly directed every federal agency to stop using Anthropic’s technology and barred any defense contractor from doing business with Anthropic at all, even on matters unrelated to defense. The Department of War then formally designated Anthropic a “supply chain risk” under 10 U.S.C. § 3252, a statute designed to address sabotage by hostile foreign actors.
What the court found
- First Amendment retaliation. The court held that Anthropic’s public statements about AI safety were protected speech, and that the government’s own directives—which repeatedly cited Anthropic’s “arrogance,” “hostile” press posture, and Silicon Valley ideology—showed that retaliation was the actual motive. Evidence that the Department of War continued negotiating with Anthropic and pursued its next-generation model even after branding the company a national security threat undercut any claim that the actions were driven by a genuine security concern.
- Due process. Anthropic had a protectible liberty interest in its reputation and its right to compete for federal contracts. Because the government imposed a permanent, government-wide bar and publicly labeled Anthropic an “adversary” without advance notice or an opportunity to respond, the court found a due process violation. Section 3252 allows limited disclosure in some cases, but the government never demonstrated the exigency needed to skip pre-deprivation process altogether.
- APA violations. Two separate failures stood out. First, the secretary’s order banning any company from doing business with both the U.S. military and Anthropic—a secondary boycott—had no statutory basis at all; the government conceded as much. Second, the underlying “supply chain risk” designation did not meet the statutory definition, which is aimed at sabotage and malicious code, not a vendor’s public negotiating posture. The court also found the government skipped required procedural steps, including a reasoned determination that “less intrusive measures” were unavailable and a risk assessment from the appropriate official.
- Section 558(b). Several agencies that terminated Anthropic’s access solely because the President directed them to do so, without independent legal authorization, were found to have issued “sanctions” under the APA.
The court declined to find the Presidential Directive itself constitutional as an ultra vires act and left standing the government’s basic right to choose its own AI vendor. The ruling does not require anyone to continue using Anthropic’s products.
Why this matters for government contractors
- “Supply chain risk” has a real legal floor. Section 3252 and similar exclusion authorities are not a general-purpose tool for punishing vendors an agency finds difficult. They require an actual, articulable risk of sabotage or malicious interference—not distrust arising from public criticism or hard-nosed contract negotiating.
- Process still applies, even in the name of national security. Contractors facing exclusion, debarment, or a “risk” designation are entitled to notice and an opportunity to respond before the government acts, absent genuine exigency. A bare invocation of national security will not excuse skipping that process.
- Public advocacy is not automatically a waivable contract term. A contractor’s public statements about the safety, legality, or ethics of its own products—even in the middle of a contract dispute—can be protected speech. Retaliating against a vendor for expressing that position, rather than for its substantive contract terms, exposes agencies to First Amendment liability.
- Secondary boycotts and blacklisting exceed ordinary contracting authority. An agency can decide not to renew or extend its own contracts. It cannot, on this record, order third parties that do business with it to sever unrelated business relationships with a disfavored vendor.
- Contemporaneous conduct is evidence. The court leaned heavily on the fact that officials continued negotiating with and discussing new business with the very company they had just labeled a threat. Contractors facing adverse action should preserve all correspondence; it may be the strongest evidence that a stated rationale is pretextual.