Inside the IDF's AI Targeting System and Its High Civilian Cost

An investigation by +972 Magazine and Local Call details how the Israel Defense Forces (IDF) relied on Lavender, an AI machine designed to generate kill lists for potential Hamas and Palestinian Islamic Jihad operatives. Unlike the previously disclosed 'Gospel' system, which targets infrastructure, Lavender marks individuals. Sources within the intelligence community describe the system as a primary driver of the military's early bombardment strategy, with officers treating AI outputs as human-vetted decisions despite a known 10% error rate.

Officers involved in the process admitted that their role was reduced to providing a cursory stamp of approval, often spending only seconds confirming a target's gender before authorizing a strike. The IDF reportedly established permissive rules of engagement, allowing up to 20 civilian deaths for each junior operative marked by the AI, and exceeding 100 civilian deaths for senior officials. Because the system primarily tracked phone usage, strikes frequently occurred in private homes, resulting in the deaths of family members if the device had been passed to a relative or an unintended user. This reliance on automated data, which marked approximately 37,000 individuals as targets in the war's initial weeks, has drawn sharp criticism from human rights experts regarding adherence to international humanitarian law and the risks of unchecked AI in modern warfare.