The Black Box 010: The UN Says the US Hit a School With Stale Targeting Data and Calls It a War Crime. The 18-Page Report Never Mentions AI. The Pentagon's Unreleased One Reportedly Does.
Bot Mutiny |
The UN found the strike on the Shajareh Tayyebeh school reckless and a war crime, on a record the United States refused to answer. Two outlets say the Pentagon's own investigators blamed overreliance on AI. That report is still in a drawer.
The Black Box 010: The UN Says the US Hit a School With Stale Targeting Data and Calls It a War Crime. The 18-Page Report Never Mentions AI. The Pentagon's Unreleased One Reportedly Does.
The UN found the strike on the Shajareh Tayyebeh school reckless and a war crime, on a record the United States refused to answer. Two outlets say the Pentagon's own investigators blamed overreliance on AI. That report is still in a drawer.
October 02, 2026
Sometime between 2013 and 2016, someone built a wall.
It ran along one side of an Islamic Revolutionary Guard Corps naval complex in Minab, a town in Hormozgan Province on Iran's southern coast, and it cut a few converted buildings off from the rest of the base. The buildings got their own gate onto a public street. By 2016 they were the Shajareh Tayyebeh Primary School, for girls and boys aged six to thirteen. Murals of children were painted on the outside walls. There was signage and a playground. Satellite imagery from December 2025 shows figures in the courtyard that look like children at play.
On the morning of February 28, 2026, in the opening hours of the US and Israeli attack on Iran, two Tomahawk missiles hit the school. The first struck the lower floor. Teachers moved the children upstairs to a prayer room. The second struck the upper floor.
The Minab prosecutor put the final toll at 156: 120 schoolchildren, 73 boys and 47 girls; 26 educators, all women; seven parents; a bus driver; a pharmacy technician from next door.
The United Nations Independent International Fact-Finding Mission on Iran released its report on the strike on September 17. It is 18 pages.
The words artificial intelligence appear in it zero times.
What the UN document says
The report is A/HRC/63/61, submitted to the Human Rights Council under resolution S-39/1. Most of it concerns the Iranian government's killing of protesters and its use of the death penalty, and it finds that Iran committed crimes against humanity. Section IV, paragraphs 76 through 116, concerns two US strikes. Section V.B, paragraphs 123 through 128, gives the legal findings.
On Minab, the Mission establishes the school's civilian character from satellite imagery, photographs, and video: the wall, the separate entrance, the murals, the classrooms. Paragraph 86: no information indicates the school was used for military purposes at the time. Paragraph 89: the school building was the intended point of impact, not collateral damage from the attack on the base next door.
On attribution, the Mission does not rely on its own investigators alone. It cites The New York Times of March 11, which reported that a preliminary US military investigation had found the United States responsible after target coordinates were generated from outdated intelligence identifying the school as part of the adjacent installation. It cites Reuters. It cites Admiral Brad Cooper, commander of US Central Command, who on May 19 confirmed that the preliminary findings "correlated" the strike on the school with a US strike. And it notes that the United States is the only party to the conflict with Tomahawk missiles.
The Mission wrote to the US government on June 16 asking for information about the strike, and on June 24 asking to meet Cooper and the director of the Pentagon's Civilian Protection Center of Excellence. It received no reply.
Paragraph 124 is the finding on precautions. The United States, when identifying and directing its attack against a school whose civilian nature is apparent, failed in its obligation to do everything feasible to verify that it was a military objective. Then the sentence that matters: updating targeting information and acting upon such information falls within the category of feasible precautions an attacking party is required to take under international humanitarian law.
Paragraph 125 is the finding on intent. The Mission concludes the United States acted with recklessness, dolus eventualis, and it says how it got there. It weighs public statements by senior US leadership, including the President and the Secretary of Defense, that dismiss binding obligations under international humanitarian law. It weighs the apparent existence of and reliance on outdated targeting data. It weighs the multiple features readily identifying the structure as a school, which were publicly available before the strike. It concludes that the failure to update the school's status in US targeting databases and to verify it before launching went beyond mere negligence.
Paragraph 126: reasonable grounds to believe the strike amounts to the war crime of making a civilian object the object of an attack.
The Mission's evidentiary standard is reasonable grounds to believe. That is an investigative threshold, not a judicial one, and the report says so. It also says, in paragraph 126, that its evidence does not include any response from the United States or any public disclosure of the US internal investigation. It reached its finding on the record it could get.
The phrase the report uses, and the phrase it does not
The report's mechanism is a database. Outdated targeting data. Targeting databases. Updating targeting information. Those phrases carry the entire finding on precautions.
Nowhere does the report say what kind of system held that data, what generated the coordinates, or whether any software recommended the target. The Mission did not have that information, because the United States did not answer its letters and has not released its investigation. The report does not speculate. Where it does not know, it stops.
That restraint is worth stating plainly, because almost everyone else writing about Minab has not shown it, and because the next document in this story does not stop where the UN did.
What the Pentagon's investigators reportedly found
The Pentagon opened a formal command investigation into the Minab strike in March. As of this writing it has not released the findings. Bloomberg, in a September 19 reconstruction of the strike, reported that the investigation has been all but complete for several months.
Bloomberg's account, attributed to US officials, is that Pentagon investigators found that flawed intelligence, outdated imagery, and an overreliance on AI contributed to the strike. The site had been identified years earlier as a military compound, and there was clear evidence it had been modified and changed purpose, but that information did not reach the authoritative targeting database.
CNN had reported the same shape in March, citing two sources briefed on the preliminary findings: CENTCOM generated the coordinates from outdated Defense Intelligence Agency information. Satellite imagery from 2013 showed the school and the base as one compound. Imagery from 2016 showed the fence. Later CNN reporting, cited by three senators in September, described senior commanders bypassing warnings in critical databases, including one powered by AI, that target intelligence was severely out of date.
Time, reporting on the UN findings, added the sentence that keeps this honest: no publicly disclosed evidence has established that an AI model independently selected the school as a target.
So the record on AI at Minab is this. Two independent outlets, citing officials and people briefed on the investigation, report that AI figured in the failure: as a database that carried stale data forward, as a warning that was bypassed, as something investigators called overreliance. No document says so. The one document that exists, the UN report, describes the failure without the word. The document that reportedly uses the word is the one the Pentagon will not release.
What the senators asked for
On September 19, Senators Mark Warner, Jack Reed, and Chris Coons wrote to Secretary Pete Hegseth and Director of National Intelligence Jay Clayton. The letter has not been posted on any of their websites; what follows is as CNN quoted it.
They wrote that recent events "have given us growing concern about the extent to which agencies under your oversight have prioritized acceleration of AI capability adoption and 'experimentation' over effective governance." They wrote that public reports of AI targeting platforms generating "spurious outputs (based on outdated input data or outright hallucination)" risk serious operational error. They asked the relevant inspectors general to be given "unrestricted access to both instances this year in which media reports have suggested significant errors in AI-enabled targeting workflows, as well as any additional instances that may have so far not been publicly reported."
Both instances. Minab is one. The other is a story CNN broke on September 18, from four unnamed sources: this spring, a special operations analyst asked a chatbot to analyze intelligence on a Chinese ship's manifest in the Middle East; the chatbot fused open-source material with classified signals intelligence and concluded the ship carried nuclear weapons components; the analyst used AI again to format that into a standard intelligence report; the military planned an intercept, armed personnel prepared to board, aircraft were airborne; and only just before the operation did officials discover the report's provenance and find it wrong. CNN could not learn what the ship carried or which chatbot was used. Every other outlet covering it is repeating CNN. The Pentagon told TMZ it does not comment on specific intelligence reports.
Bot Mutiny reports that account as CNN's, and nothing more. It is in this piece because the senators put it in their letter, next to Minab, under one heading: outdated input data or outright hallucination.
The memo that set the clocks
Seven weeks before Minab, on January 9, Hegseth signed a six-page memorandum titled the Artificial Intelligence Acceleration Strategy. Defense One and Forecast International each read it in full.
Per their accounts, the memo directs the Chief Digital and Artificial Intelligence Office to ensure vendors can deploy their latest models inside the department within 30 days of public release, and calls that a primary procurement criterion. It orders benchmarks for model objectivity within 90 days. It orders standard "any lawful use" language in every AI procurement contract within 180 days. It sets monthly demonstrations on seven pace-setting projects, one of which is agentic AI for battle management "from campaign planning to kill chain execution," and another of which is described as turning intelligence into weapons in hours rather than years. Its final section is headed: Clarifying "Responsible AI" at the Department of War: Out with Utopian Idealism, In with Hard-Nosed Realism. Defense One reported that the memo omits any mention of ethical use of AI and does not reference the department's earlier AI ethics principles.
The memo's named success metrics, per Forecast International, are AI system usage and mission impact.
Neither outlet reported a deadline in the memo for refreshing targeting data, a standard for when a model's output must be verified by a person, or a rule for withdrawing a model that gets things wrong. [AR: open the PDF at media.defense.gov and confirm or correct this before publishing.]
Hegseth announced the strategy at SpaceX on January 12. The press release quotes him: "We will unleash experimentation, eliminate bureaucratic barriers, focus our investments and demonstrate the execution approach needed to ensure we lead in military AI." The senators put the word experimentation in quotation marks in their letter.
The innocent reading
It deserves its full weight.
The UN Mission never entered Iran, never spoke to the United States, and worked from public reporting, satellite imagery, and Iranian state media footage. Its finding on intent leans on political speeches by the President and the Secretary of Defense, which is an unusual basis for dolus eventualis, and a September 19 post on Opinio Juris titled "The Iran IIFFM's Unconvincing War Crimes Analysis" argues the point. Reasonable grounds to believe is a low bar by design.
Stale target data has killed civilians in US strikes for decades without any AI in the loop. The 2021 Kabul strike is the standard example. If a targeting database held a 2013 entry and nobody refreshed it, that is a human and institutional failure whatever software sat on top. Bloomberg's officials named three contributors, and AI was one of three.
And the ship incident, whatever it was, ended with nobody boarding and nobody hurt. Someone checked.
Why the second reading holds anyway
It holds because of what the two documents say when set side by side, and it does not require believing anything about the ship.
The UN report's operative finding is that updating targeting data is a feasible precaution the United States was required to take and did not. The January memo, as reported by two outlets, contains clocks for adopting models and a benchmark for their politics, and no clock for updating the data those models run on. The one thing the UN says was legally required is the one thing the strategy did not schedule.
The memo's success metrics are usage and mission impact. Minab produced high usage and, in the department's own terms, a completed mission. Nothing in those two metrics would have flagged it. The senators' letter now asks for the category of event the strategy was not designed to count.
The Pentagon's silence is itself part of the record. It did not answer the UN's June letters. It has not released an investigation that Bloomberg's sources say has been finished for months. On March 31, CENTCOM said US forces do not target civilians. In May, Cooper told Congress the military was investigating only one case of possible US-caused civilian deaths, this one. The senators' September letter asks for "any additional instances that may have so far not been publicly reported." Three senators do not write that sentence about a department they believe has disclosed everything.
What the instrument cannot see
A targeting database does not know how old it is. It holds an entry until someone changes it. The UN report's finding is that changing it was required. Bloomberg's officials say the change existed somewhere and never reached the authoritative database. Whatever system read that database on the morning of February 28 read a school as a base, because that is what the entry said.
The January memo measures how much the department's AI is used and whether missions succeed. By both measures, Minab was not a failure. It is invisible to the instrument. That is the structural finding, and it stands whether or not a single sentence of the ship story is true.
The record
What the documents establish:
- A/HRC/63/61, released September 17, 2026: the Shajareh Tayyebeh Primary School was a clearly identifiable civilian school; the building was the intended point of impact; 156 people were killed by the prosecutor's count, including 120 children and 26 women educators; the United States carried out the strike; updating targeting data was a required feasible precaution and was not taken; the Mission has reasonable grounds to believe this was the war crime of making a civilian object the object of an attack. The Mission asked the United States for information on June 16 and June 24 and got no reply.
- The January 9, 2026 memorandum, as read by Defense One and Forecast International: 30-day model deployment as a primary procurement criterion; 90-day objectivity benchmarks; 180-day "any lawful use" contract language; success metrics of usage and mission impact.
- The Warner, Reed, and Coons letter of September 19, as quoted by CNN: a request for inspector general access to two named incidents and any unreported ones.
What they do not establish:
- That AI selected the school. No document says it. The UN report does not use the term. Bloomberg and CNN report that officials and briefed sources describe overreliance on AI and a bypassed AI-powered warning; those are attributed accounts of an unreleased investigation.
- The contents of the Pentagon's investigation. It has not been released.
- That the ship incident occurred as described. One outlet, four unnamed sources, no document, no official confirmation.
- The full text of the senators' letter. Only CNN's quotations are public.
- That the January memo contains no data-currency or verification standard. Two careful readers reported none. [AR: confirm against the PDF, then delete this bracket.]
The UN report ends its section on the United States with one demand, in paragraph 129: take immediate steps to disclose the outcomes of its internal inquiries into strikes on civilians. As of publication, 216 days after the strike, the only people who know what the Pentagon's investigators wrote about AI at Minab are the Pentagon and the reporters its officials talk to.
Sources: United Nations Human Rights Council, Report of the Independent International Fact-Finding Mission on the Islamic Republic of Iran, A/HRC/63/61, advance edited version dated September 21, 2026, paragraphs 76 to 98 and 123 to 129. Department of War memorandum, "Artificial Intelligence Strategy for the Department of War," January 9, 2026, as reported by Patrick Tucker, Defense One, January 13, 2026, and Vincent Carchidi, Forecast International Defense Security Monitor, January 22, 2026. Department of War press release, January 12, 2026. Bloomberg, "Inside US Military 'Kill Chain' That Destroyed an Iranian School," September 19, 2026. CNN, "US military had close call after using AI for false intelligence report, sources say," September 18, 2026, and "Democrats call for investigation into faulty AI-assisted intel report," September 19, 2026. Time, "U.S. May Have Committed War Crimes in Iran, U.N. Report Says," September 18, 2026. Jessica Dorsey, Opinio Juris, September 18, 2026.