U.S. Military: AI Misfires and Strikes School in Iran

The U.S. Military: AI Misidentifies Target and Strikes a School in Iran

The U.S. military used AI to select thousands of targets but missed a note indicating that one of them was a school.

An investigation into a missile strike on an Iranian school highlights serious shortcomings in the U.S. military's targeting infrastructure. AI is supposed to address these gaps.

A missed note from an intelligence analyst and systems that did not communicate with each other: according to a report from the Los Angeles Times, these are the two central failures investigators uncovered while examining a missile strike on an Iranian school. The attack, which occurred in late February, killed around 120 children. This strike took place during a war where the U.S. military, according to previous reports, used AI on a large scale for target selection for the first time. The Claude model from Anthropic was integrated into Palantir's Maven Smart system and suggested around 1,000 targets on the first day.

Years before the strike, an analyst had noticed changes at a site in the city of Minab, in southeastern Iran. The U.S. had previously classified the building as an Iranian naval military facility. In the meantime, it had become an elementary school.

A Note That No One Ever Saw

The analyst reported the changes in 2019 using a digital intelligence tool, according to the LA Times. The critical issue was that the tool was not linked to the official target database that the U.S. military uses to develop strike targets. The information never reached the commanders. The building was reviewed multiple times, but no one updated the database. According to the New York Times, the images used were seven years old.

At least two intelligence databases were never connected to the authorized target database, the LA Times reports. In Syria, targeting data from the mid-2010s was sometimes 10 to 20 years old. At the center is a database called MIDB, built in the 1980s, which still relies heavily on manual entries. It is supposed to be replaced by an automated system called MARS, but the transition is years behind schedule. The U.S. Government Accountability Office reported long-standing deficiencies in the system as early as 2020.

This aging infrastructure stands in stark contrast to the speed of AI elsewhere. A report from the WSJ indicated that the number of targets hit in the early days exceeded 3,000 and warned that the oversight mechanisms for human review of lethal decisions were underfunded. Even at that time, U.S. investigators believed that U.S. forces were likely responsible for the strike on the school, a conclusion that the LA Times report now supports with specific technical failures.

AI Is Supposed to Fix What Failing Databases Cannot

Some targeting experts hope that connecting digital systems and adding more AI will reduce errors in the future, the LA Times reports. An automated check against public services like Google Maps could flag anomalies for human review. The Pentagon has moved in this direction after the report, unveiling an agentic AI initiative.

The Defense Intelligence Agency, which oversees both MIDB and MARS, did not directly address the flaws or the transition delays when contacted by Bloomberg. A spokesperson referred generally to the thorough analysis conducted by assigned analysts.

The Pentagon's AI Pioneer Sounds the Alarm

According to the current U.S. targeting doctrine, military commanders decide whether to prioritize and strike a target. They must distinguish military objects from civilian objects. There is also an optional process called target verification that checks the accuracy of the underlying intelligence. A former senior intelligence official told the LA Times that it would be unthinkable for a commander to skip this step during strikes on the first day of a new campaign. Centcom reviewed the targets before operations against Iran, but it remains unclear whether the optional verification process was initiated.

The sharpest criticism in the report comes from a striking source. Jack Shanahan, a retired Air Force general and the first director of the Joint Artificial Intelligence Center established in 2018, led the Project Maven AI program. This makes him one of the architects of AI adoption in the U.S. military, the same military that now relies on this very Maven system. At the time, Shanahan predicted that AI would play a central role in any potential conflict between the U.S. and China, and that in 20 years, algorithms would compete with each other.

Shanahan told the LA Times that there is no excuse for a command not to verify the accuracy of its intelligence. He described targeting itself as a moribund professional field that has withered over two decades while the military focused on counterterrorism. As early as 2017, he said, he could barely find people to fill these roles.