A report by transparency watchdog Airwars, titled *Anatomy of an AI Kill Chain*, examines how AI and machine learning systems drive modern military targeting decisions across six stages, from surveillance to post-strike assessment, with humans involved in only a minority of steps. The report uses a fictitious but realistic kill chain to highlight the compounding errors possible at each stage, including flawed translations, misidentification by computer vision, and automation bias that makes it difficult for humans to meaningfully intervene. Co-author Sophia Goodfriend argues that the fundamental flaws in these machine learning systems are not merely teething problems, but inherent limitations that militaries cannot rely on when making life-and-death decisions.
