US appeals court upholds Pentagon decision to blacklist Anthropic as supply chain risk

Appeals court ruling in Washington

A three-judge panel of the U.S. Court of Appeals for the District of Columbia Circuit ruled on September 25, 2026, that the Pentagon acted within the law when designating artificial intelligence company Anthropic as a supply chain risk. The 51-page decision rejected Anthropic's challenge against the Defense Department, leaving in place a ban that prevents the startup from receiving military contracts and stops defense contractors from using its Claude model for government defense work. Writing for the majority, Judge Gregory Katsas stated that the executive branch holds constitutional authority over national security procurement decisions.

In our Republic, it is up to the president and the secretary of defense to determine how to weigh the risks between an AI that is too constrained and one that has no limits.

Judge Katsas determined that the exclusion stemmed from contractual terms rather than an attempt to penalize Anthropic for expressing its viewpoints. In her dissent, Judge Karen Henderson cautioned that the ruling forces private suppliers into an untenable position. She argued that commercial defense providers will now face a choice between accepting the defense secretary's requirements or risking designation as a threat to national security.

Origins of the military contract dispute

The standoff began in February 2026 during negotiations regarding a $200 million contract to deploy Anthropic technology on classified systems. Defense Secretary Pete Hegseth demanded the removal of software restrictions that prevent Claude from assisting in mass domestic surveillance of American citizens and autonomous lethal weapons. Anthropic refused to lift those safeguards, maintaining a corporate policy in favor of slowing rapid artificial intelligence development to manage catastrophic risks.

The U.S. armed forces had previously utilized Claude across sensitive networks, including during the January 2026 military operation that captured and deposed Venezuelan leader Nicolas Maduro. Following Anthropic's refusal in February, the Pentagon placed the company on its supply chain risk registry in March, an administrative designation previously reserved for foreign enterprises associated with hostile nations. The military subsequently transferred its contracts to competitor firms, reaching an agreement with xAI in February, signing OpenAI on the day of Anthropic's exclusion in March, and concluding deals with Google and Microsoft in May.

Legal divergence and financial impact

The decision in Washington creates a direct conflict with an August 2026 ruling issued by Judge Rita Lin in the U.S. District Court for the Northern District of California. In that separate lawsuit, filed on March 9, Judge Lin struck down a broader executive directive ordering all federal civilian agencies to stop using Claude, finding the Trump administration had unlawfully retaliated against the firm for protected speech.

The hollow invocation of national security is not a blank check to punish those who criticize the government and retaliate.

The regulatory confrontation comes while Anthropic, valued at nearly $1 trillion, prepares for an initial public offering in autumn 2026. The company stated that the military blacklist caused billions of dollars in lost business and reputational harm, compounded by a three-week period in June 2026 when the government imposed export controls on its advanced models. Relations with the White House have remained strained, with President Donald Trump dismissing warnings about AI dangers from Anthropic chief executive Dario Amodei as a hoax. Anthropic confirmed it is reviewing the ruling and evaluating options, including requesting a rehearing before the full 11-judge panel of the D.C. Circuit.