How AI suggested 1,000 targets yet missed the school note
investigators found the us military missed a 2019 analyst flag about an iranian school because ai systems didn't talk to legacy databases, highlighting lethal gaps in targeting infrastructure.
investigators probing a strike that killed 120 children in minab, iran, found that while the us military used ai to select over 3,000 targets, the system failed because a 2019 analyst note flagging the site as a school was never entered into the official target database. the failure stems from a disconnect between modern ai tools, including anthropic's claude used via palantir's maven system, and legacy databases like the midb, which still rely on manual data entry from the 1980s. the pentagon's new strategy involves 'agentic ai' initiatives designed to automatically cross-reference targeting data with public sources like google maps to prevent similar oversights. this case reveals the stark danger of pairing high-velocity ai targeting with stagnant infrastructure, where the speed of lethal decisions outpaces the verification of fundamental intelligence.