Pentagon Can Blacklist Anthropic Over Claude Features, Court Says

Pentagon blacklist Anthropic Claude features court ruling

On September 25, 2026, a federal appeals court upheld the Department of Defense’s decision to blacklist Anthropic’s technology. The ruling came from the US Court of Appeals for the District of Columbia Circuit, where a two-to-one majority affirmed that the Trump administration acted within its authority when it barred federal agencies and defense contractors from using Anthropic’s AI products. The court said the dispute centers on how powerful AI systems should be used in national security settings.

Department of Defense supply chain risk AI model constraints court

The majority warned that overly constrained models could shut down unexpectedly during critical missions, jeopardizing military operations. At the same time, it noted that completely unconstrained systems might hallucinate targets and recommend inappropriate lethal force. The judges said officials must weigh these competing dangers when deciding how to deploy AI in defense contexts. They concluded that the secretary of defense did not exceed any limits set by law or the Constitution in reaching his determination.

The panel consisted of Judge Gregory Katsas and Judge Neomi Rao, both appointed during the first Trump administration, who formed the majority. Judge Karen Henderson, appointed by President George H. W. Bush, dissented. Henderson argued that the statutory language at issue was aimed at deliberate sabotage or espionage by hostile actors, not at a contractor’s transparent refusal to enable certain features. She believed the broader interpretation risked punishing legitimate business decisions.

This decision differs from an earlier ruling by a district court in the Northern District of California. That judge found the blacklisting unlawful because Anthropic did not meet the narrow definition of a supply-chain risk under 10 U.S.C. § 3252, which requires malicious intent by an adversary. The appeals court, however, reviewed the matter under 41 U.S.C. § 4713, a provision that does not demand a bad motive and grants the DC Circuit exclusive jurisdiction over procurement actions made under that section. The court pointed out that the language of section 4713 covers “any person” who might interfere with a product’s design, operation or data flow, which includes Anthropic’s decision to keep certain Claude capabilities disabled for military use.

Because the broader statute does not require proof of harmful intent, the majority said the department’s concern that Anthropic might manipulate Claude to prevent it from performing lawful national-security functions qualified as a supply-chain risk. The judges emphasized that the administration’s action was permissible even though Anthropic acted without malicious intent.

Anthropic responds to blacklisting claims retaliation impact on developers

Anthropic responded by saying it respectfully disagrees with the court’s view and remains confident in its position. A spokesperson noted that another federal court had already deemed the government’s parallel designation unlawful and indicated the company is considering further appeals, either en banc review before the full appeals court or a petition to the Supreme Court. Despite the legal fight, Commerce Secretary Howard Lutnick recently remarked that the Trump administration and Anthropic have mended their relationship and are now “in tune.”

Anthropic has also argued that the blacklisting amounts to retaliation after it refused to lift restrictions on using its models for lethal autonomous warfare and mass surveillance of Americans. The district court judge in California agreed with that claim, finding the government’s action to be an unlawful punitive measure. The appeals court majority did not address the retaliation allegation directly, focusing instead on the statutory interpretation of supply-chain risk.

For developers and organizations that build or rely on AI systems, the case highlights how government procurement rules can affect access to advanced models, especially when those models are subject to use-case limitations. It underscores the importance of understanding which legal provisions govern contracting decisions and how differing interpretations can lead to opposite outcomes in different courts. Stakeholders may want to monitor further appeals and consider how licensing terms and feature enablement strategies could influence eligibility for future government contracts.

Frequently asked questions

Why did the Pentagon blacklist Anthropic's technology?

The Department of Defense barred federal agencies and contractors from using Anthropic’s AI products because it was concerned that Anthropic might manipulate its Claude models to prevent them from performing lawful national-security functions, which the appeals court treated as a supply-chain risk under 41 U.S.C. § 4713.

Which court upheld the blacklisting and what was the vote?

The US Court of Appeals for the District of Columbia Circuit upheld the blacklisting with a two-to-one majority, with Judges Gregory Katsas and Neomi Rao in the majority and Judge Karen Henderson dissenting.

What statutes did the appeals court rely on, and how did they differ from the district court’s approach?

The appeals court reviewed the case under 41 U.S.C. § 4713, which does not require proof of harmful intent and gives the DC Circuit exclusive jurisdiction, whereas the district court had applied 10 U.S.C. § 3252, which demands malicious intent by an adversary and found the blacklisting unlawful.

Source: Ars Technica