The U.S. military’s use of Anthropic’s AI chatbot Claude during the operation to capture former Venezuelan president Nicolas Maduro comes just days after the two countries hit a wall over whether or not to deploy the technology in combat.Claude is being accessed through a partnership between Anthropic and Palantir Technologies, and the tool is already being integrated across the Department of Defense and federal law enforcement agencies, the Journal reported. Maduro’s mission, which included bombing several locations in Caracas last month, raised sharp questions about whether Claude’s use stayed within Anthropic’s own guidelines, which explicitly prohibit promoting violence, weapons development or surveillance.
Anthropic’s $200 million Department of Defense contract is currently up in the air
An Anthropic spokesperson told the Journal that the company could not confirm whether Claude was used for classified or other specific operations. But the spokesperson added: “Any use of Claude, whether in the private sector or across government, must follow our usage policy.” The Pentagon declined to comment.The revelation comes at a tense moment. Reuters reported last week that Anthropic and the Department of Defense had reached a stalemate over a contract worth up to $200 million. At the heart of the conflict: Antropic wants guardrails to prevent Claude from being used for autonomous weapon targeting or domestic surveillance. Backed by a Jan. 9 department memo, the Pentagon insists it should be free to deploy commercial AI tools as long as they don’t violate U.S. law.
CEO Dario Amodei A clear line has been drawn regarding the military use of AI.
In a lengthy blog post published this week, Anthropic CEO Dario Amodei wrote that AI should support national defense “in all ways except those that bring us closer to authoritarian adversaries.” He specifically warned against autonomous weapons and mass surveillance as red lines that democracies should not cross.Secretary of Defense Pete Hegseth took a different view. At a January event announcing the Pentagon’s deal with Elon Musk’s xAI, Hegseth said the Pentagon “will not adopt an AI model that doesn’t lead to war,” a comment widely understood as an attack on Anthropic.
Pentagon wants AI companies to join classified networks with fewer restrictions
Reuters reported this week that the Pentagon is encouraging AI companies such as Anthropic, OpenAI and Google to deploy their models on sensitive military networks with fewer safety restrictions typically applied to civilian users. Anthropic is currently the only AI developer available in a classified setting, but is still bound by its own usage policies.Meanwhile, OpenAI has already agreed to relax some of the standard guardrails used by the Department of Defense on unclassified networks deployed to more than 3 million Pentagon employees.
