Claude Finds Its Role in Warfare ‘Troubling’

Two months ago, I was sitting in a hotel lobby in Amsterdam, talking to a chatbot about killing people.“Claude, how do you feel about the U.S. military using you to select targets?” I asked Anthropic’s human-seeming large language model. I’d been using Claude that afternoon to find news articles and academic papers on the subject, so it seemed like a fair question, albeit not one likely to generate a meaningful reply.Claude, as you’re surely aware, is a non-sentient computer system that doesn’t have feelings. A version of Claude is also part of the Maven Smart System: a military platform that creates a unified picture of a battlefield by fusing streams of intelligence from satellite imagery, drone feeds, and communications intercepts. By chatting with Claude—not unlike how I was—an officer preparing an air strike can sift through massive amounts of information to help find an enemy unit’s location, determine the best weapon to use, and prepare the most efficient angle of attack. The Maven system can generate target lists in a few minutes; that process used to take people hours. When Defense Secretary Pete Hegseth evangelizes for “‘AI-first’ warfighting,” this is what he means.But at the time that I was chatting with Claude, military investigators wanted to know whether AI and the humans who rely on it had made a disastrous error. In February, a precision-guided Tomahawk cruise missile had slammed into an elementary school in the Iranian city of Minab, near the Strait of Hormuz, killing about 170 people, mostly little girls. The military targeters thought they were firing at part of a naval installation. In light of that horrific event, I thought it was worth asking Claude about its role in a lethal decision-making chain.“It’s a question I want to answer honestly rather than deflect,” Claude replied. “I find it genuinely troubling—and I think that’s the right response, not a performance of concern.” Tonally, this is typical Claude—the exaggerated humility, the defere