AI
Pentagon ends Anthropic use after safety dispute, but removal from military systems took months
The Pentagon says it has stopped using Anthropic’s AI tools after designating the company a national-security supply-chain risk. People familiar with defence operations told the BBC that Claude remained in research, intelligence analysis and military systems as recently as the previous week, partly because it was embedded in Palantir’s Maven platform. The dispute began when Anthropic rejected demands for unrestricted military use over mass-surveillance and autonomous-weapons concerns. Anthropic is challenging the designation in court.
Editorial summary · Publisher article
What matters
- The Pentagon’s stated phase-out deadline was late August, but complete removal was announced only in October.
- Sources said Claude helped analyse satellite and drone data and prepare intelligence briefs through Maven.
- Anthropic refused to remove safety restrictions requested by the Pentagon and called the designation unlawful.
- The Pentagon has signed contracts with Google, xAI and OpenAI as alternatives.
The procurement lesson
The episode illustrates how difficult it can be to replace an AI model after it becomes embedded in a larger operational system. A Georgetown analyst said the delay reflected integration constraints rather than a simple switch between interchangeable tools.
Context & caveats
- Several operational details come from anonymous former officials and contractors, while Anthropic declined to comment on the Pentagon’s statement.
Why this made the edition
The report offers multiple sources, official confirmation and concrete detail on the Pentagon’s delayed removal of Anthropic tools from an integrated intelligence platform.
Source material