Pentagon Used Anthropic's Claude AI in Military Operation
The U.S. military reportedly used Anthropic's Claude large language model for planning and real-time support, including target selection, in the recent capture of Venezuelan leader Nicolás Maduro. Following the AI's use in a capacity that appears to violate the company's own ethical policies, the Pentagon is now reviewing Anthropic as a potential “supply chain risk.” The deployment highlights the rapid integration of agentic AI into tactical military operations.
- The operation in which Claude was used was the capture of Venezuelan President Nicolás Maduro in January; U.S. special operations forces struck targets across Caracas, and Venezuela's defense ministry reported 83 casualties. - Anthropic’s public usage policy explicitly prohibits using its models for “military and warfare” or activities with a high risk of physical harm, including weapons development. The current dispute centers on Anthropic's refusal to remove safeguards that prevent Claude from being used for mass surveillance of U.S. citizens or in fully autonomous weapon systems. - The AI model was accessed through a partnership with Palantir Technologies, a major defense contractor that integrates third-party AI like Claude into its platforms for government clients. After the raid, an Anthropic employee reportedly contacted Palantir to ask if their software was used, which concerned Pentagon officials about the company's reliability for future missions. - In response to the disagreement, the Pentagon is now threatening to designate Anthropic a "supply chain risk," a severe measure typically used for foreign adversaries. This would require any company doing business with the Department of Defense to certify that they do not use Anthropic's technology in their workflows. - This standoff is particularly fraught because Anthropic's Claude is currently the only large language model of its kind available on the military's classified networks, where officials have described it as highly effective. The Pentagon has a contract with Anthropic worth up to $200 million. - The Department of Defense is simultaneously negotiating with Google, OpenAI, and xAI to gain access to their models for "all lawful purposes," with those companies reportedly showing more flexibility than Anthropic in removing safeguards for military use cases on unclassified systems. - This incident occurs as the U.S. Army is actively developing large language models for tasks like target recognition and pairing weapons to targets, aiming to evolve them into "reasoning models" that can replicate the skillsets of a field artillery warrant officer. - The Pentagon's broader AI adoption effort is guided by its 2023 "Data, Analytics, and Artificial Intelligence Adoption Strategy," which prioritizes gaining "decision advantage" through the rapid and agile implementation of AI. The strategy is executed by the Chief Digital and Artificial Intelligence Office (CDAO), which aims to accelerate the deployment of AI capabilities at scale.
