> ## Content Index
> Fetch the complete content index at: https://www.counterpremise.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Humans in the Loop: AI Wasn’t the Failure in Minab
- URL: https://www.counterpremise.com/humans-in-the-loop-ai-wasnt-the-failure-in-minab/
- Published: 2026-08-23T22:35:30.000Z
- Updated: 2026-09-01T06:34:50.000Z
- Description: A US strike killed 123 children in a primary school. The missile appears to have hit the target it was given. There were humans in the loop. The harder questions start there.
- Author: Aymar Pirzada

It was Saturday morning in Minab, during Ramadan. As American and Israeli strikes began elsewhere in Iran, staff at Shajareh Tayyebeh primary school started calling parents to collect their children. Shortly after 10 a.m., Iranian authorities announced that schools nationwide would close.

One father arrived for his ten-year-old son. Two younger relatives were still waiting for their own father. He offered to take them too, but they stayed behind.

He left with his son. About ten minutes later, he heard the explosions.

At least 157 people were killed at the school. [Researchers have since identified 123 of them as children](https://airwars.org/citation/ap-cites-airwars-minab-documentation/?ref=counterpremise.com), none older than thirteen.

Responsibility for the February 28 attack was initially contested. Iran blamed the United States and Israel. Israel denied involvement. American officials said the United States would never intentionally target a school and opened an investigation. President Donald Trump suggested Iran itself might have been responsible.

Five months later, the picture is considerably clearer. Amnesty International and Human Rights Watch concluded that [the school itself was directly struck with guided weapons](https://www.amnesty.org/en/latest/news/2026/03/usa-iran-those-responsible-for-deadly-and-unlawful-us-strike-on-school-that-killed-over-100-children-must-be-held-accountable/?ref=counterpremise.com). 

Bellingcat and Forensic Architecture have used satellite imagery, verified videos and other open-source evidence to [reconstruct the attack](https://www.bellingcat.com/news/2026/03/08/video-shows-us-tomahawk-missile-strike-next-to-girls-school-in-iran/?ref=counterpremise.com) and its multiple impacts. 

Subsequent US reporting has [traced the strike further upstream](https://apnews.com/article/2a134a5c74d80db763db4c3eb6d0d847?ref=counterpremise.com), into the intelligence and targeting process that put the school in the path of a U.S. weapon. The emerging American explanation has focused on outdated intelligence.

That may explain part of what happened. It does not yet explain enough.

## **A school with a military history**

Shajareh Tayyebeh had [real connections to the Islamic Revolutionary Guard Corps](https://www.reuters.com/investigations/bombed-iranian-girls-school-had-vivid-website-yearslong-online-presence-2026-03-12/?ref=counterpremise.com). Its site had once formed part of the adjoining military complex. The school belonged to an IRGC-affiliated educational network, and some pupils reportedly had parents serving in the Guard. Others were local children, including members of Iran’s Sunni Baluch minority, a community that has itself suffered repression by the Iranian state.

Whatever the politics of the institution or its parents, the children were civilians.

Reuters reconstructed [a civilian footprint that was hardly obscure](https://www.reuters.com/investigations/bombed-iranian-girls-school-had-vivid-website-yearslong-online-presence-2026-03-12/?ref=counterpremise.com). The school maintained a public website; photographs showed children studying and playing there; online directories identified it as a school. Satellite imagery showed that it had been physically separated from the neighbouring military facilities years earlier, with its own walls and entrances.

Associated Press reporting on the Pentagon investigation later added an extraordinary detail from inside the American intelligence system: [a US analyst had identified the building as a school years before the attack](https://apnews.com/article/2a134a5c74d80db763db4c3eb6d0d847?ref=counterpremise.com). That information was apparently not sufficiently shared across the intelligence and military staffs involved in targeting.

The building apparently existed elsewhere in the American targeting system under a different identity. Early reporting described outdated intelligence associated with the adjoining military site; subsequent reporting has [described the building in US targeting information as a factory](https://www.washingtonpost.com/national-security/2026/03/11/us-strike-iran-elementary-school-ai-target-list/?ref=counterpremise.com), while other accounts pointed to a weapons-related target at the location.

A precision weapon did not have to malfunction for a catastrophe to occur. It only had to be given a target based on information that was wrong.

## **The missile didn’t miss**

Minab has inevitably become part of a wider argument about artificial intelligence and warfare. Modern militaries increasingly use AI-assisted systems to process intelligence and accelerate target development. The Iran war has intensified questions already raised by Gaza and Ukraine about what happens when militaries can identify and prosecute targets at unprecedented speed.

The scale of that acceleration is difficult to overstate. In the first 38 days of Operation Epic Fury, US forces struck [more than 13,000 targets](https://www.armscontrol.org/act/2026-05/news/ai-plays-major-role-war-iran?ref=counterpremise.com) in Iran. The Pentagon’s Maven Smart System fused intelligence from satellites, drones, radar and communications, helped identify and rank targets, and could generate new target lists within minutes. US officials presented that speed as a battlefield advantage. Critics warned that the same compression could leave human reviewers with less time to scrutinize or double-check what the system put in front of them. 

There is, however, no public evidence that an artificial intelligence independently selected Shajareh Tayyebeh for destruction. Nor is there evidence that a weapon went astray.

[Sky News and Forensic Architecture later reconstructed the attack in three dimensions](https://news.sky.com/story/revealed-a-catalogue-of-failures-and-the-devastating-aftermath-of-iran-school-bombing-13564022?ref=counterpremise.com) using satellite imagery, videos, photographs and witness testimony. Their analysis placed a precision strike on the school itself, including the area where children had been sheltering.

The failure appears to have occurred before launch.

Someone decides which intelligence is authoritative and how recently a target needs to be checked. Someone designs the process for handling contradictory information held by different parts of the same intelligence system. Technology can make that process faster. It cannot make the underlying information true.

Calling Minab a database error risks making an institutional failure sound like a technical glitch.

CNN reported in July, citing three people familiar with the targeting process, that [senior US commanders bypassed warnings in critical databases that intelligence on some potential Iranian targets was severely outdated](https://transcripts.cnn.com/show/cg/date/2026-07-28/segment/02?ref=counterpremise.com). Two of those sources said the decision was made for expediency as officials rushed to produce targets at the beginning of the war. Some of the intelligence underlying the strike list was reportedly more than a decade old. Minab was among the strikes affected.

If that reporting is correct, stale intelligence was not simply an error waiting to be discovered. Its age had become a known risk that people in the targeting process were prepared to accept.

That changes the question.

## **How much uncertainty was acceptable?**

Months before the war, Defense Secretary Pete Hegseth told senior American military leaders that US forces should use [“overwhelming and punishing violence”](https://www.war.gov/serve-from-netstorage/News/Transcripts/Transcript/Article/4318689/secretary-of-war-pete-hegseth-addresses-general-and-flag-officers-at-quantico-v/index.html?ref=counterpremise.com) and be empowered to “intimidate, demoralize, hunt and kill” their enemies. He called for “maximum lethality” and attacked restrictive rules of engagement.

Ten days into the Iran campaign, Hegseth described the coming day as [the most intense yet — “the most fighters, the most bombers, the most strikes”](https://www.cbsnews.com/news/pete-hegseth-dan-caine-news-briefing-pentagon-iran-war/?ref=counterpremise.com) — and told reporters that American intelligence was “more refined and better than ever.”

During the same campaign, Hegseth celebrated the removal of what he called [“stupid rules of engagement”](https://www.realclearpolitics.com/video/2026/03/02/hegseth%5Fto%5Fthe%5Fmedia%5Foutlets%5Fand%5Fpolitical%5Fleft%5Fscreaming%5Fendless%5Fwars%5Fstop%5Fthis%5Fis%5Fnot%5Fendless.html?ref=counterpremise.com). American troops, he said, were trained to kill the enemy and “break their will.”

There is no public evidence that Hegseth or another senior official ordered anyone to attack the children at Minab, and his words cannot tell us why Shajareh Tayyebeh was struck. But they are relevant to the priorities being communicated down the chain of command.

If commanders were accepting greater uncertainty about intelligence to generate targets quickly, then the emphasis placed on speed, lethality and fewer constraints deserves scrutiny alongside the database failure. The question is not whether aggressive rhetoric caused one particular strike. It is what kinds of error the institution was prepared to tolerate in pursuit of operational tempo.

Minab may still prove to have been an extraordinary intelligence mistake despite reasonable safeguards. The July reporting raises a more troubling possibility: safeguards were bypassed because waiting for better information would have slowed the production of targets.

Those possibilities are not equivalent.

Five months later, the Pentagon has not told us which one occurred.

## **Where the public record ends**

Witnesses have alleged that the school was struck again after the initial attack, when survivors had been moved and parents or rescuers had arrived. Bellingcat independently established [multiple impacts and at least two waves of strikes around the school and neighbouring IRGC compound](https://www.bellingcat.com/news/2026/03/27/new-videos-reveal-more-details-about-iran-girls-school-strike/?ref=counterpremise.com). Its reconstruction did not establish the full sequence described by those witnesses or show that American forces knowingly targeted survivors.

That allegation should remain an allegation. The US military’s operational records should help establish what happened.

The Pentagon began investigating Minab shortly after the strike and later elevated its inquiry into a formal investigation. Its findings remain unpublished. In July, more than [two dozen US senators called for an unclassified accounting](https://www.hickenlooper.senate.gov/press%5Freleases/hickenlooper-colleagues-demand-pentagon-release-investigation-of-deadly-iran-school-strike/?ref=counterpremise.com) of what happened and what was being done to prevent a recurrence.

CNN subsequently reported an additional concern. [The first two stages of the military review had been completed](https://transcripts.cnn.com/show/cnr/date/2026-07-29/segment/19?ref=counterpremise.com), but a broader review that would ordinarily examine what happened and its implications had not been ordered as of early July. The Pentagon said the investigation remained ongoing.

The government therefore possesses evidence unavailable to the public: the target package, intelligence underlying it, approval chain and operational records. Those materials could establish why information identifying Shajareh Tayyebeh as a school failed to prevent it from becoming a target, what warnings commanders received and what safeguards were applied.

They could also tell us what has changed since.

## **The humans in the loop**

The debate over AI-enabled warfare often comes back to the reassurance that humans remain “in the loop.” Minab shows how little that phrase tells us on its own.

Anthropic CEO Dario Amodei [offered precisely that insistence](https://www.amnesty.org/en/latest/news/2026/06/usa-four-months-after-horrific-minab-school-airstrike-accountability-delayed/?ref=counterpremise.com) after the strike. Asked about Claude’s possible role, he acknowledged that Anthropic did not know exactly how its models had been used. He called the attack “a really terrible thing to happen,” but added that [it did not violate Anthropic’s red lines](https://scribehawk.com/transcript/inside-the-mind-of-anthropic-ceo-dario-amodei-the-circuit-extended-interview??ref=counterpremise.com): Claude could assist, while a human retained the final decision. “A human made that final call, not Claude,” he said.

There were humans in this loop: people assembled the intelligence, reviewed potential targets and determined what uncertainty was acceptable. Technology can make those decisions faster. It can also distribute them across databases, analysts, software and command structures until responsibility becomes harder to locate.

But attributing Minab to technology would be strangely comforting. Machines malfunction. Software has bugs. Databases become outdated. Each suggests a problem for which a technical fix might be found.

A functioning primary school became something American forces believed they could attack. Information contradicting that belief apparently existed elsewhere inside the same intelligence apparatus. Later reporting suggests warnings about the age of some of the intelligence were known and overridden.

The children killed at Shajareh Tayyebeh had no place in that decision-making chain. They depended on the people inside it to distinguish their school from a military target.

What happened inside that chain matters beyond Minab.

As militaries become more dependent on AI-assisted targeting, the difficult question will not only be whether machines are permitted to make lethal decisions autonomously. It will also be what happens when humans make those decisions through systems so complex and so fast that human agency becomes obscured by the machinery around them.

A database did not decide how old intelligence could be before it became too dangerous to use, or how much uncertainty was acceptable in the rush to generate targets. Nor did the missile decide whether the information attached to its coordinates justified destroying what stood there.

Those were human decisions.

We do not yet know which of them led to Shajareh Tayyebeh. The United States possesses evidence that could tell us considerably more, and it should release it.

Until then, describing Minab as a failure of outdated intelligence may tell us where something went wrong. It does not tell us who accepted the risk of being wrong — or who should answer when that risk is borne by 123 dead children.