Ethan Carter
In February, two Tomahawk missiles struck Shajarah Tayyebeh Elementary School in Minab, Iran, on the first day of the Iran war. The attack resulted in the tragic loss of over 150 lives, including at least 123 children.
Initial reports indicated issues with outdated targeting data and raised concerns about the potential involvement of artificial intelligence in the error. A recent Bloomberg investigation has provided further details, citing insiders from an unreleased internal Pentagon review. These officials revealed that some personnel at the Pentagon were aware within hours that the U.S. had hit the school and described a series of preventable mistakes, notably an excessive reliance on an AI tool developed by Palantir.
Sources from U.S. Central Command indicated that some team members depended heavily on the artificial intelligence within the Maven Smart System, an AI-driven data integration and targeting platform designed by Palantir to enhance intelligence analysis and decision-making speed. Over the past year, the Defense Department has adopted this software tool as a key element of military operations. Notably, several former senior military officials are now employed by Palantir, including individuals with high-level clearances at Centcom’s headquarters in Tampa.
According to officials, some users anticipated that Maven would identify outdated records or inconsistencies in the intelligence gathered for potential targets. it remains unclear why they expected the system to perform these functions. The Minab location was misidentified as an Islamic Revolutionary Guard Corps facility based on old data, which was inputted into Maven along with other sites and emerged as a recommended target for the first day of operations. The process of compiling target lists, which previously required hours, was reduced to mere minutes.
A spokesperson for Palantir stated that the company “is not responsible for the underlying data nor identifying intelligence deficiencies” and emphasized that there is no indication that their software contributed to the error. Sources familiar with Palantir’s contracts with the Pentagon noted that the government retains primary responsibility for the quality of the information fed into Maven. Following the attack, Palantir implemented features that “re-review underlying intelligence to identify factors that would disqualify a target and flag inconsistencies and inaccuracies that human review may have missed,” according to an insider. This new functionality has reportedly already detected some anomalies.
In March, over 120 House Democrats sent a letter to Defense Secretary Pete Hegseth, inquiring about the role of Maven and other AI tools in identifying the Minab site. The Pentagon has not publicly responded, citing the ongoing investigation. Officials informed Bloomberg that the complete report has been nearly finished for several months. A Pentagon spokesperson confirmed that the incident remains under investigation and declined to provide further comments.
The overreliance on Maven was just one of three critical failures in the targeting process outlined to Bloomberg. Other factors included inaccurate intelligence and outdated satellite imagery.
U.S. databases had classified the Minab compound as a military site for years, yet commercial satellite images depicted a different reality. By 2017, construction efforts had completed walls and separate entrances that isolated the school from the adjacent military base. A satellite image from 2018 showcased brightly colored walls, a soccer field, assembly areas, and playground markings.
In June, Bloomberg reported that an analyst had noticed these changes as early as 2019 and had documented their observations in a system that was not linked to the primary military intelligence database used for targeting. Unfortunately, these notes never reached the personnel responsible for assembling the strike package. Human intelligence from southern Iran was limited; however, the school maintained its own website and was easily found on Google Maps.
The site continued to be labeled as an IRGC facility as the Trump administration called for a significant initial assault. More than 1,000 Iranian targets were struck within the first 24 hours, and sources indicated that this rapid pace limited the time available for further confirmations.
Staffing for civilian harm mitigation teams across the U.S. Department of Defense has decreased by approximately 90% in recent years, dwindling to fewer than 20 personnel overall. Centcom’s team reduced from 10 members to just one. None of these team members reviewed the Minab site prior to the missile launch.
A separate investigation conducted by the United Nations’ Independent International Fact-Finding Mission on Iran concluded this week that there are reasonable grounds to believe the United States committed the war crime of executing an indiscriminate attack. The report stated that the U.S. “failed in its obligation to do everything feasible to verify” whether the school was a military objective and that the failure “went beyond mere negligence.” It claimed that the United States “directed the strikes at the building of the school while being aware of a substantial risk of striking a civilian object and acting recklessly as regards the possibility that this would happen.”
Despite these findings, Washington has yet to publicly accept responsibility. In March, President Trump stated, “No, in my opinion, based on what I’ve seen, that was done by Iran. We think it was done by Iran, because they’re very inaccurate, as you know, with their munitions.” In a more recent interview with Fox News, Trump remarked, “I don’t think anybody’s going to ever be able to say what happened there.” A senior official from the administration told Bloomberg, “The United States does not target civilians,” and the Defense Department has refrained from commenting on the UN report.
Errors related to overreliance on AI are emerging as a recurring issue as this technology becomes increasingly integrated into military and law enforcement practices. In a different incident reported this week, a special operations analyst utilized a chatbot that incorrectly suggested a Chinese-flagged ship in the Middle East was transporting nuclear weapon components destined for Iran. The U.S. military was poised to board the vessel but ultimately withdrew at the last moment.
The combined application of facial recognition and AI technology in law enforcement has also led to disastrous outcomes. In Florida, an automated facial recognition match resulted in the wrongful arrest of a man who had never visited the state and was 400 miles away at work, causing him months in jail and the loss of his job and home. In another instance, a man was arrested based on a match despite being over 300 miles away at the time of the alleged crime. Charges in both cases were eventually dropped, but such incidents are part of a much longer list of wrongful arrests across the nation based on flawed software evaluations.
While individuals across various fields may occasionally over-rely on AI while understanding its practical capabilities, the stakes are significantly higher in military and law enforcement contexts. Here, human error or the acceptance of automated outputs without verifying the underlying data can lead to catastrophic consequences. Given this reality, it is crucial to focus on how humans utilize this technology, rather than solely on the theoretical potential for uncontrolled power.
Daniel Mercer
