General David Petraeus told a panel at the NDTV World Summit that the U.S. has already fired 13,000 strikes using an AI‑driven system in Operation Epic Fury over Iran, a number that dwarfs any previous campaign. The first day alone saw 1,000 hits, a tempo the Pentagon says was enabled by AI that cuts decision time from days to seconds.
The system, dubbed Maven Smart System, is built by Palantir and powered by Anthropic’s Claude language model. It ingests real‑time geospatial data from the National Geospatial‑Intelligence Agency, runs it through machine‑learning models, and assigns confidence scores to potential targets, but hands the final strike order to a human commander.
The same architecture was trialed on the Russian‑Ukrainian front and in Israel’s Gaza offensive, where IDF units used AI tools like Lavender and Hasbora to flag buildings and human targets. Those deployments reportedly identified 37,000 potential targets, but critics note that the systems were stripped of advanced targeting to avoid lethal decisions.
An Iranian school strike on February 28 that killed 150 people has reignited the debate over AI‑assisted targeting. While investigations have not conclusively linked Palantir’s system to the choice of target, they highlight gaps in intelligence vetting and human oversight.
The Pentagon’s approval of the Maven system for future missions means that the U.S. will rely on AI to identify, evaluate, and strike targets with unprecedented speed, but it also obliges the Department of Defense to establish robust accountability protocols for civilian casualties.