US appeals court upholds Pentagon's exclusion of Anthropic

A US appeals court has upheld the Pentagon's exclusion of Anthropic from contracts as a supply-chain risk, backing the military in a dispute over limits on the use of its Claude AI system. [1] [5]

The US Court of Appeals for the District of Columbia Circuit rejected Anthropic's legal challenges in a 2–1 ruling on 25 September 2026. A separate California judgment blocking wider restrictions on the company remains in force. [1] [5]

Limits on military use

The Pentagon had sought contractual permission to deploy Claude for all lawful uses. Anthropic wanted to retain exclusions for mass domestic surveillance and fully autonomous weapons. [1] [3]

In a February statement, Anthropic chief executive Dario Amodei argued that mass domestic surveillance was incompatible with democratic values. He also said current frontier AI systems were not reliable enough to power weapons that select and engage targets without human involvement, while acknowledging that fully autonomous weapons might become important for national defence. [3]

The court's majority found sufficient grounds for the Pentagon to regard Claude as potentially unreliable for military operations because of its built-in restrictions and the unresolved contract dispute. It rejected Anthropic's free-speech and due-process challenges, describing the central dispute as contractual. [1] [5]

Judge Karen LeCraft Henderson dissented. She favoured a narrower interpretation of the supply-chain law, arguing that a contractor's open enforcement of restrictions was not the kind of risk that justified exclusion, ABC News reported. [5]

Wider restrictions remain blocked

The California case concerned a separate designation under a different law. On 27 August, US District Judge Rita Lin ruled that the designation was unlawful and that the government had violated Anthropic's free-speech and due-process rights. [2]

Lin also blocked President Donald Trump's government-wide ban on Anthropic and Defence Secretary Pete Hegseth's prohibition on military contractors doing any business with the company, Reuters reported. ABC News reported that the California ruling remained in effect on 25 September. [4] [5]

Lin found that the government's measures were intended to punish Anthropic for criticising it, rather than resting on an articulable basis to expect sabotage. Her judgment also recorded that Anthropic had no backdoor access to its technology once deployed in a national-security system, and that the government conceded the technology itself was no riskier than other black-box AI models. [2] [5]