They struck on the first day of the Iran war. I read the list of targets and then the message: a school in Minab had been hit — more than 150 dead, at least 123 of them children. The air filled with a question that has haunted me since: how did a Tomahawk find a playground instead of a target?
I want you to hold that moment — the sound of a missile, the silence after — as we trace how people, software and old images combined into a catastrophe. I have reviewed the reporting, the leaked summaries, and spoke with sources; you will see where decisions stacked and why they mattered.
An analyst logged the site’s change in 2019; Too Much Trust in AI
An analyst noted new walls, painted courtyards and a soccer pitch years before the strike. Yet, inside U.S. Central Command, components of the Maven Smart System produced a recommended target list and the Minab compound surfaced as an IRGC site. I watched how fast an algorithm can become the de facto referee when humans step back.
Officials told Bloomberg that some Centcom staff expected Maven — the Palantir-built intelligence integration platform — to catch stale or conflicting inputs. The Pentagon has leaned on Palantir software this past year to collapse hours of target analysis into minutes. Palantir says it does not control the underlying intelligence; the government keeps responsibility for data quality. Still, several former senior officers with ties to Palantir worked at Centcom, and users told investigators they treated the system as authoritative.
The system acted like a lighthouse that shone false beams — bright guidance where the maps were wrong. After the strike, engineers added checks to force re-review of raw intelligence and flag inconsistencies; those features reportedly caught anomalies in later work.
Did Palantir’s AI cause the Minab school strike?
No single actor owns the answer. The unreleased Pentagon probe describes a cascade: outdated database entries, ignored analyst notes stored outside the primary targeting feed, rushed decision cycles, and heavy reliance on a tool that organized the data. Palantir’s software recommended candidates; it did not fire Tomahawks. Still, when humans assume algorithms are complete, responsibility blurs.
Commercial images from 2018 showed a school; Outdated imagery of a military site that had become a school
Satellite photos from 2018 displayed painted walls, playground markings and assembly lines — unmistakable school features. Those images had never replaced the older military designation in primary U.S. databases.
Bloomberg’s reporting found an analyst flagged the site change as early as 2019, but those notes sat in a disconnected system and never reached the strike package builders. Google Maps even listed the school and the site had its own website. Yet the strike list moved forward amid an administration-directed overwhelming opening assault that hit more than 1,000 Iranian targets in 24 hours. The blitz compressed time for extra checks.
What did Pentagon investigators find about the kill chain?
The internal review reportedly found three linked failures: bad, outdated intelligence; unchecked automated recommendations from the Palantir system; and degraded civilian-harm mitigation staffing inside the Department of Defense. Centcom’s civilian harm team dwindled from ten people to one. No one from those teams reviewed Minab before the missiles launched.
House Democrats asked about Maven in March; UN investigators call it a war crime
More than 120 House Democrats demanded answers from Defense Secretary Pete Hegseth about Maven’s role. The Pentagon has answered only by saying the strike remains under investigation.
The UN’s Independent International Fact-Finding Mission on Iran concluded there are reasonable grounds to believe the attack was an indiscriminate strike and that the United States “failed in its obligation to do everything feasible to verify” the target. The mission states the failure rose above negligence and that the U.S. acted recklessly given a substantial risk of hitting civilians. Washington has not publicly accepted responsibility; President Trump initially told reporters he believed Iran was to blame and later said nobody might ever say what happened.
Is the U.S. responsible for the Minab strike?
Responsibility is both legal and moral: the UN report frames a legal argument; the Pentagon probe aims to explain what went wrong operationally. Politically, the administration denies targeting civilians; legally, investigators are pointing to failures that created a known risk. That gap between political denial and forensic accounting is where accountability will be fought.
Staff cuts and hallucinating chatbots; A familiar pattern of errors in military and law enforcement use of AI
Civilian-harm teams were reduced by roughly 90% across the Department of Defense. The human checks that used to catch mistakes largely disappeared. I have seen similar threads in law enforcement cases: an automated facial recognition match sent an innocent man to jail, another match arrested a man who was hundreds of miles away.
In separate recent reporting, a special operations analyst used a chatbot that hallucinated cargo of nuclear components on a Chinese-flagged ship, nearly triggering a boarding. These are not failures of theory; they are failures of practice — people accepting automated outputs without verifying the raw facts. The intelligence stack began to behave like a house of cards: one misplaced piece and the whole pattern collapses.
Where lives and reputations hang on a few lines of data, software becomes a blunt instrument unless paired with people and time to test its inferences.
I want you to imagine the families in Minab watching the sky. If the Pentagon’s internal review confirms preventable mistakes tied to outdated images, disconnected analyst notes and overreliance on Palantir’s Maven, who should answer for the children?