SecurityResearch

US Military: AI target selection fails due to legacy infrastructure – Iran school bombing exposes system breakdown

An investigation into the February airstrike on an Iranian school that killed approximately 120 children reveals that while the US military deployed AI for target selection, critical information never reached commanders because separate databases were not connected.

In detail

  • Anthropic's Claude was embedded in Palantir's Maven Smart System and suggested roughly 1,000 targets on day one; over 3,000 targets were hit in the first days.
  • An analyst documented in 2019 that a building in Minab had been converted from an Iranian military facility to an elementary school – but this information was not linked to the official target database.
  • The central target database MIDB dates from the 1980s, still relies heavily on manual input, and was supposed to be replaced by the MARS system – this transition is years behind schedule.
  • At least two intelligence databases have never been connected to the authoritative target database; in Syria, target imagery was sometimes 10–20 years old.

Why it matters

The failure demonstrates that AI speed without robust data management and system integration leads to catastrophic errors – a cautionary tale for anyone deploying AI in critical processes where data quality and governance cannot keep pace.

For you In your AI projects, verify that your data sources are actually connected and that outdated information is automatically filtered – fast AI decisions on poor data are worse than slow ones.

← All news

Summaries are generated automatically and link to the original source.