Pentagon review: overreliance on Palantir’s Maven AI contributed to the US strike on a school in Minab, Iran
Partly confirmed
Importance: major (4 of 5)Status
- Claim
Partly confirmed
- Our reporting
- Medium confidence
- Importance
- Major (4 of 5)
- Last verified
- 30 September 2026
Your AI and this story
- GPT-6.2 Astra110 days after its cutoff
- Claude Opus 5.580 days after its cutoff
- Gemini 4 Hydrogen171 days after its cutoff
- Grok 4.7110 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 80 days before it.
Key facts
- Strike: Feb 28, 2026, opening day of the Iran war; two Tomahawk missiles hit the Shajarah Tayyebeh Elementary School in Minab; more than 150 killed, at least 123 children (Bloomberg via Gizmodo/IBTimes)
- The site was catalogued as an IRGC facility from outdated data; commercial imagery had shown a school (painted walls, soccer pitch, playground) since at least 2017–2018
- Review finding: CENTCOM personnel relied heavily on AI in Palantir’s Maven Smart System, expecting it to flag outdated or inconsistent intelligence; Maven compressed targeting work from hours to minutes but did not catch the misclassification
- Civilian-harm mitigation staff across DoD fell ~90% to fewer than 20 people; CENTCOM’s team went from 10 people to 1, and no civilian-harm official reviewed the Minab site before launch
- Some Pentagon personnel knew within hours that the US had hit a school
- Palantir: it ‘is not responsible for the underlying data nor identifying intelligence deficiencies’; it later added features to re-examine underlying intelligence (Gizmodo)
- Pentagon declined public comment citing the ongoing investigation; Bloomberg (Sept 22) reported the military has since modified its AI-assisted combat targeting
- LLM role unclear: Gizmodo says LLMs such as Claude were not involved in this strike; Wikipedia notes Maven incorporates Anthropic’s Claude
What happened
Bloomberg’s investigation, based on Pentagon officials’ accounts of an unreleased internal review, found no single catastrophic decision behind the Minab strike but an accumulation of smaller failures. The school sat on an old target list as an IRGC site. Maven, the Palantir targeting platform used by US Central Command, recommended it as a day-one target, and staff trusted the system to surface stale or contradictory intelligence, which it did not. At the same time the teams that check targets for civilian harm had been cut to almost nothing.
The Bloomberg graphic was published around Sept 18 (Wikipedia’s date); Gizmodo’s write-up is dated Sept 20 and the HN thread Sept 22. The review itself was not published, so the findings are as reported by Bloomberg (confidence medium).
Why it matters
It is the clearest documented case of automation bias in AI-assisted targeting causing mass civilian deaths. It arrived in the same weeks as the fight over the Pentagon’s “supply chain risk” label for Anthropic and the weakening of the UN text on autonomous weapons.
Sources
6 sources from 5 sites. Numbers match the chips in the text.
6 sources: 4 press, 2 reactions
Press
- Bloomberg: Inside the US military ‘kill chain’ that destroyed an Iranian schoolbloomberg.com, press
- Bloomberg: US military modifies AI, combat targeting after Minab school strike (Sept 22)bloomberg.com, press
- Gizmodo: Pentagon investigators say overreliance on Palantir AI tech contributed to strike that killed 123 Iranian childrengizmodo.com, press
- IBTimes UK: Pentagon review of AI failures in Iran school strikeibtimes.co.uk, press
Reactions
- Wikipedia: 2026 Minab school attacken.wikipedia.org, discussion
- Hacker News discussion (972 points)news.ycombinator.com, discussion
Changes
- Filed (HN 972 points)