ARCUHN

ARC Lab

No. 03 ·

Lees in het Nederlands: ENNL

They saw the school. The database never did.

ARC Analysis · Anomaly, Reframe, Create

In short

AAnomaly

On 28 February 2026, a US missile hit an elementary school in Minab, Iran. More than 120 children were killed 1, 2. In 2019, a US analyst had noticed the building was now a school. He wrote it down, in a tool that was not linked to the targeting database 2, 3. Investigators named an "overreliance on artificial-intelligence technology" among the causes 1.

RReframe

The knowledge was there. It sat in the wrong place, and everyone trusted that someone, or something, else would check. AI did not create that gap. It made it fast.

CCreate: questions for your team

  1. When someone notices that something has changed, does it reach the system that decides?
  2. What do we expect our tools to check, that they never promised to?

Rethink the premise.

Go deeper The full piece, with all sources

AAnomaly

On 28 February 2026, a US missile hit an elementary school in Minab, Iran. More than 120 children were killed 1, 2. According to Bloomberg, which spoke to officials involved in the Pentagon's internal inquiry, it was not one catastrophic decision but "an accumulation of smaller ones" 1.

The site had once been part of a military compound. In 2019, a US analyst noticed it had changed. He wrote it down, in a digital tool that was not linked to the official targeting database. Over the following years, the same building was reviewed several more times, without anyone updating that database 2, 3. In the first day of the war, more than 1,000 targets had to be handled. Investigators named time pressure, data gaps, cuts in civilian-protection staff and an "overreliance on artificial-intelligence technology" 1.

Research on automation bias shows a pattern. People who work with automated systems start to rely on what the system flags, and miss what it does not. Those who felt personally accountable for how they used the system checked more, and made fewer errors 4.

RReframe

The headlines say AI killed children. The harder truth is quieter. The knowledge was there. Someone saw it. Someone wrote it down. It never reached the system that decided. Each step made sense on its own: a note in one tool, a database in another, a system that was trusted to be right. Together, the note and the decision never met. AI does not fix old data. It makes old data fast.

This is a story about a war. Underneath, it is a story about data, and that part is not only military.

CCreate: questions for your team

  1. When someone notices that something has changed, does it reach the system that decides?
  2. Who owns our old data?
  3. What do we expect our tools to check, that they never promised to?

Rethink the premise.

Sources

  1. Inside US Military 'Kill Chain' That Destroyed an Iranian SchoolBloomberg (Bartenstein & Karra) · 2026 · Investigation, 18 Sep, based on officials of the internal inquiryStrong · paywall; original read via screenshots
  2. U.S. analyst's missed remark surfaced in Iran school strikeBloomberg (Manson), via Los Angeles Times · 2026 · Investigation, 28 JunStrong · paywall; original read via screenshots
  3. The US military used AI to pick thousands of targets but missed a note saying one was a schoolThe Decoder · 2026 · News, summarising Bloomberg · https://the-decoder.com/the-us-military-used-ai-to-pick-thousands-of-targets-but-missed-a-note-saying-one-was-a-school/Medium · secondary, open access
  4. Automation bias: decision making and performance in high-tech cockpitsMosier, Skitka et al., NASA · 1997 · Experiments · https://ntrs.nasa.gov/citations/20020043179Strong