Pentagon AI Overreliance Killed 123 Children in Iran Strike

According to a Bloomberg investigation published Friday, September 18 and amplified by Gizmodo, U.S. Central Command's overreliance on Palantir's Maven Smart System contributed directly to a February 2026 Tomahawk missile strike that killed more than 150 people, including at least 123 children, at Shajarah Tayyebeh Elementary School in Minab, Iran. The strike occurred on the opening day of the Iran war, and officials involved in an unreleased internal Pentagon review told Bloomberg that some personnel knew within hours the U.S. had hit the school. The site had been cataloged as an Islamic Revolutionary Guard Corps facility because of outdated targeting data, then fed into Maven with other candidates and returned as a recommended day-one target. Earlier reporting already pointed to outdated targeting data and raised questions about whether artificial intelligence had a role; the Bloomberg investigation now confirms that concern. Maven Smart System, built by Palantir, is an AI-powered data integration and targeting platform that the Department of Defense has made a cornerstone of military operations over the past year. The tool condenses target-list work that once took hours into minutes. According to the Bloomberg account, personnel inside Centcom leaned too hard on the artificial intelligence inside Maven, with some users expecting the system to flag stale records or contradictions in the intelligence assembled for potential targets. It remains unclear why they expected that capability, because Maven was not designed to perform data quality assurance or identify intelligence deficiencies. A Palantir spokesperson told Bloomberg that the company "is not responsible for the underlying data nor identifying intelligence deficiencies" and that there is no evidence its software was at fault. The company's defense does not dispute that outdated data flowed through Maven; it disputes that flagging it was Palantir's job. The investigation also highlights a concentration of former senior officers now working for Palantir, including some with high-level clearances at Centcom's Tampa headquarters. This revolving door is not new in defense contracting, but it takes on added weight when the contractor's software is implicated in a mass-casualty event. It raises questions about whether Centcom's trust in Maven was shaped by technical merit, institutional familiarity, or the presence of former colleagues. Even if no individual acted improperly, the perception of conflict undermines the legitimacy of AI-enabled targeting. Officials described a cascade of preventable failures, suggesting that human oversight may have been degraded by the very efficiency Maven promised. The story drew intense technical scrutiny on Hacker News, with the Bloomberg piece scoring 125 points and 61 comments. The market and regulatory implications for Palantir are significant. As a publicly traded company, PLTR faces tail risk from congressional hearings, potential revisions to military AI procurement, and demands for explainability and audit trails. Investors will weigh the near-term certainty of defense contracts against longer-term reputational and regulatory costs. For the broader defense-tech sector, the Minab strike is likely to accelerate calls for mandatory human-on-the-loop checks, automated staleness flags, and independent verification of targeting data. The tragedy also provides a concrete case study for AI safety researchers and policymakers: automation bias is not theoretical when operators assume a system will catch what it was never designed to see. Looking ahead, the unreleased Pentagon review may become public, forcing a more granular accounting of what Maven did and did not do. The U.S. military will likely impose stricter data validation protocols before target recommendations can proceed, and Palantir may face contractual language assigning responsibility for data quality. Allies and adversaries are watching. China and Russia will cite the strike to discredit U.S. claims about responsible military AI, even as they pursue similar systems. The core lesson is that AI can compress time but cannot eliminate the need for human judgment. In Minab, the time saved by Maven came at the cost of 123 children, and that equation will shape defense AI governance for years.