US strike on Iranian school caused by human error, not AI: Semafor
An investigation reveals human error, not AI, caused a US strike on a school in Iran, raising concerns over military targeting systems and civilian safety.
-
Portraits of school children from the Shajareh Tayyebeh Elementary School in Minab, who were killed in a US strike, are displayed during a press conference by Iranian Ambassador to Tunisia Massoud Hosseinian, in Tunis, Tunisia, March 12, 2026 (AP)
The US attack that killed more than 150 schoolgirls at an elementary school in Iran was the result of human error, not artificial intelligence, an investigation by Semafor revealed.
The attack on Shajareh Tayyebeh Elementary School, which was carried out during the first wave of strikes on Iran amid the US-Israeli war, has drawn intense scrutiny over how such a high-casualty mistake could occur. While public debate has increasingly focused on the role of AI in modern warfare, those with insight into the bombing campaign say the failure lies squarely with the human systems used to identify and approve targets.
According to Semafor, US military targeting relies on extensive databases compiled well in advance of conflict, built by thousands of analysts who process satellite imagery and intelligence inputs. In this case, officials failed to detect subtle but critical changes on the ground, while intelligence analysts either missed or failed to incorporate publicly available information indicating that a functioning school was located within the supposed IRGC compound.
The oversight proved catastrophic. Analysts did not update or flag the site as civilian, leaving it on target lists despite evidence that could have prevented the strike.
Those familiar with the process say the error reflects a broader limitation of human-led intelligence systems, which can struggle to keep pace with evolving conditions and large volumes of data. By contrast, AI systems, despite their own well-documented flaws, are capable of processing far greater amounts of information and cross-referencing sources in ways that might have identified the school before the strike.
“Finding the right balance between humans and machines will be a crucial component of future training,” Jack Shanahan, a former senior US military official and founding director of the Pentagon’s Project Maven AI program, told Semafor.
AI selected the target, humans failed to review
New military technology is allowing the US to generate and process far larger volumes of targets at far greater speed than in previous wars, but the systems remain heavily dependent on human verification, a gap that proved critical in the strike on an Iranian school, according to Semafor.
The targeting process now draws on a fusion of satellite imagery, radar, human intelligence, and other data streams; however, artificial intelligence is primarily used to identify unknown or moving targets, rather than fixed sites that have already been logged in military databases.
In the case of Iran, the Islamic Revolution Guard Corps’ Sayyid al-Shuhada complex in Minab would have been treated as a “deliberate target", selected in advance through intelligence gathering and satellite analysis, former military officials told Semafor. Such sites are typically part of long-standing target lists compiled over decades and maintained by US Central Command, with oversight from military lawyers.
Had the location been identified as a school, it would have been removed from the list, military experts told Semafor.
According to Semafor, the process also includes a final layer of human review in the 24 to 48 hours before a strike, when analysts are expected to confirm that targets remain valid.
If discrepancies are identified, standard practice calls for further examination using advanced imaging tools capable of comparing satellite photos in detail to detect changes on the ground. It remains unclear whether that step was carried out.
A review of publicly available information, cited by Semafor, showed clear indications that a school was operating within the compound. Satellite imagery also revealed structural changes, including new walls and a separate entrance, consistent with civilian use. But without direct comparison to older intelligence, human reviewers relying on outdated data may have had little reason to question the site’s classification as a military target.
Decision-making remains in human hands
The Pentagon’s targeting workflow is increasingly powered by Palantir’s Maven Smart System, a platform that allows military planners to rapidly assemble strike plans by integrating satellite imagery, intelligence feeds, and summaries generated by AI tools like Anthropic’s Claude, according to Semafor.
The system can also suggest potential targets and help prioritize how to strike them, significantly speeding up what was once a far slower, manual process. But despite its capabilities, decision-making remains in human hands, with no indication that the military intends to remove people from the “kill chain", Semafor reported.
Still, experts warn that the growing reliance on such systems carries risks. “There will be a risk of becoming overly reliant on machine outputs and short-circuiting critical human review,” Jack Shanahan told Semafor.
For now, Semafor noted, the more immediate concern is not autonomous AI decision-making, but whether human reviewers can keep pace with the speed and scale at which these systems generate targets, and maintain sufficient scrutiny to avoid civilian harm.