On the morning of Feb. 28, 2026, the first day of the US war with Iran, two US Tomahawk missiles struck the Shajarah Tayyebeh elementary school in Minab, in southern Iran. More than 150 people were killed, at least 123 of them children, and Bloomberg's investigation of the strike calls it the deadliest American military targeting error of the 21st century in terms of child casualties. Officials involved in the Pentagon's internal investigation, which has not been released, told Bloomberg the strike came from an accumulation of small failures: flawed intelligence, satellite imagery years out of date, and personnel who leaned too hard on an artificial intelligence system the military had built into its targeting. That system was credited with enabling the pace of the day's attack: about 1,000 targets in 24 hours.
A school that stayed on the target list
Minab is a town in southern Iran. The school was built almost a decade ago on land that had once been part of a military compound used by a naval unit of the Islamic Revolutionary Guard Corps, a military force separate from Iran's regular army. A base can become a school, and this one did.
The US targeting record did not follow. The site stayed catalogued as an IRGC facility. According to a bipartisan group of senators pressing the Pentagon to release the investigation, investigators reportedly found that several failures led the school to be targeted in error: in 2019 an intelligence analyst noted the site had changed from a naval facility to a school but entered the note into a digital system that was not connected to the authoritative targeting database, and the officials planning the attack used imagery that had not been updated in several years and did not show a school at the site.
Bloomberg reports the school was then fed into Maven along with other potential targets and emerged as one of the recommended targets for the first day of the war.
What Maven is, and what it does
Maven Smart System is an AI-enabled battle-management program made by Palantir Technologies. According to Arms Control Today, the Pentagon used it to identify top-priority targets and help choose which weapons to use against them, and Bloomberg reports it fuses more than 150 different data inputs into one picture.
Software of this kind does not fire anything. It collects what sensors, databases and reports say about a battlefield, merges them into a single display, and can rank targets and propose options within it. The practical change is that a person works from one screen rather than many files, and can move from spotting a target to acting on it faster. Cameron Stanley, who heads the Pentagon's digital and artificial intelligence office, told Arms Control Today that Maven lets a commander go from identifying a target to choosing a course of action and acting on it within a single system, a change he called revolutionary.
That magazine also reports that Anthropic's Claude AI was merged with Maven technology in late 2024 to give military users what it described as enhanced targeting options. What the officials involved in the investigation told Bloomberg is that some personnel relied too much on the AI embedded in the system, alongside intelligence that was wrong and imagery that was old.
A thousand targets in a day
The opening day was the system's showcase. Bloomberg reports the US hit 1,000 targets in a surprise attack on Feb. 28, and that Stanley said Maven enabled it. The campaign did not slow: an April 8 White House accounting counted more than 13,000 targets struck in the first 38 days, including more than 2,000 command-and-control targets, 1,500 air defense targets and 1,450 industrial base targets.
A thousand targets in a day is about forty an hour, one every 90 seconds. Whatever checks per target existed before this war, at that rate each one gets a smaller slice of a person's attention. That arithmetic is ours. The finding in the record is narrower: outdated imagery, a database that still called the site a military facility, and a machine-assembled picture some planners trusted further than the evidence under it warranted.
The team that checked for civilians
US Central Command, which runs American military operations in the Middle East, had a team whose job was to look at planned strikes on behalf of the civilians under them. People with knowledge of the cuts told Bloomberg the team was cut from 10 people to 1 under Defense Secretary Pete Hegseth, and officials involved in the inquiry say no member of it reviewed the Minab site before the strike.
What the UN found, and how Washington answered
On Sept. 17, 2026, the UN Independent International Fact-Finding Mission on Iran published findings that there were reasonable grounds to believe the United States committed a war crime by launching indiscriminate attacks, naming the Tomahawk strike on the school. According to Antara's account of the report, it covered a second February strike as well, on a sports complex and residential area in Lamerd, also in southern Iran, and the mission's conclusions rest on the evidence available to it, which did not include the US internal investigation.
The mission's sharpest words were about process. It found the US had failed in its obligation to do everything feasible to verify the school was a military target, a failure the report described as going beyond simple negligence.
Washington rejected the finding. The State Department said the administration had withdrawn from the UN Human Rights Council over what it described as persistent anti-American rhetoric and antisemitism, and a State Department official told the Associated Press that the administration does not give credibility to the report's findings.
What the military changed
Bloomberg's Sept. 22 report, which had not been published before, describes three areas of work at Central Command: refining the processes and workflows for refreshing and vetting planned targets, adding new open-source data feeds to better track where civilians are, and making dozens of upgrades to Maven Smart System. Palantir also built capabilities into Maven that re-review the underlying intelligence, to flag factors that would disqualify a target and inconsistencies that human review may have missed, according to a person familiar with the matter.
The three areas named are process, data and software. Bloomberg's account does not describe a change to the size of the civilian-harm team.
The Pentagon's investigation began in March. Officials told Bloomberg it has been all but complete for several months and it has still not been released: as of a July 15 letter from Senator Elissa Slotkin, Senator Kirsten Gillibrand and colleagues, Congress had not received the findings. The department says the incident remains under investigation and declined to comment further. What is publicly known about how a school came to be on an American target list therefore rests on unnamed officials describing a report no one outside the government has read.