According to a U.S. Department of Defense official cited by the BBC, the Pentagon has formally ceased using Anthropic’s Claude large language model. This marks Claude’s official exit from the Maven Smart System, operated by Palantir, ending approximately two years of military AI collaboration between the two sides.
The dispute originated in February this year. The Department of Defense demanded that Anthropic remove safety guardrails in the Claude model related to autonomous weapons and mass surveillance, allowing unrestricted military access. Anthropic refused. Defense Secretary Pete Hegseth subsequently designated Anthropic a national security supply-chain risk and set a maximum six-month transition period, originally scheduled for completion by the end of August.
Anthropic later filed a federal lawsuit challenging the decision as unlawful. A lower court had issued a preliminary injunction favorable to Anthropic, but on September 25 the U.S. Court of Appeals for the D.C. Circuit ruled 2–1 in favor of the Department of Defense, finding that continued integration of Claude posed a statutory national-security risk.
Because Claude was deeply embedded in the Maven intelligence platform—handling satellite imagery, drone footage, multi-source intelligence fusion, target screening, and intelligence summarization—technical removal proved difficult, delaying full decommissioning until October. Sources familiar with the matter told the BBC that as recently as last week, U.S. forces were still using Claude to process large volumes of intelligence data during military operations against Iran.
The Maven system, developed under Palantir’s lead, is architected to support multiple models. With Claude’s departure, the Department of Defense has activated alternative solutions and is gradually transitioning to large models from OpenAI, Google, xAI, and other providers. In mid-September, Under Secretary of Defense for Research and Engineering Emil Michael stated that approximately 90% of classified AI workloads had already been migrated, with Maven-related components transitioned “months ago.”
Industry analysts note that the episode highlights a core tension in military AI: the armed forces seek models free of safety and ethical constraints, while AI companies insist on maintaining protective guardrails. This parting of ways has also underscored for U.S. domestic AI firms the compliance and ethical risks attached to government military contracts.
(Compiled from BBC, DefenseScoop, and related public reporting)