A missile strike on an Iranian school has put the US military's targeting systems under new scrutiny. The late-February attack killed an estimated 120 children, and investigators found that a warning about the site's real use had existed for years but never reached the commanders who relied on the official targeting record.
The case is not only about one missed update. It shows a larger tension inside modern military targeting: AI systems can suggest targets at high speed, but they still depend on databases, human review and connected information pipelines that may be old, incomplete or disconnected.
A school remained in the target record
According to a Los Angeles Times report, the site in Minab in southeastern Iran had once been classified by the US as an Iranian military naval facility. Years before the strike, an analyst noticed that the building had changed use. By then, it had become an elementary school.
The analyst flagged those changes in 2019 using a digital intelligence tool. The central failure was that this tool was not connected to the official target database used by the US military to develop strike targets. The note did not reach commanders, and the official record was not updated.
The building was reviewed multiple times. Even so, the database continued to carry old information. According to the New York Times, the imagery used was seven years old.
That sequence matters because targeting depends on the assumption that the record in front of a commander reflects the current facts on the ground. In this case, the source describes a system where a relevant human observation existed but remained trapped outside the authoritative process.
AI moved fast, but the infrastructure lagged
The strike occurred during a war in which the US military, according to earlier reports, used AI at scale for target selection for the first time. Anthropic's Claude model was embedded in Palantir's Maven Smart System and suggested roughly 1,000 targets on day one.
A Wall Street Journal report put the number of targets hit in the first days at over 3,000. It also warned that oversight mechanisms for human review of lethal decisions were underfunded. Investigators already considered American forces likely responsible for the school strike, and the Los Angeles Times report added specific technical failures behind that conclusion.
The contrast is stark. AI targeting can generate or prioritize options quickly, but that speed does not solve the problem of stale information. If an AI-enabled system pulls from an official target database that has not absorbed critical updates, the technology can accelerate a flawed process rather than correct it.
The source identifies several weak points in the surrounding infrastructure:
- At least two intelligence databases have never been connected to the authoritative target database.
- In Syria, target data in the mid-2010s was sometimes 10 or 20 years old.
- The central database, MIDB, was built in the 1980s and still relies heavily on manual input.
- MIDB is supposed to be replaced by MARS, but that transition is years behind schedule.
- The US Government Accountability Office flagged long-standing deficiencies in the system back in 2020.
These details point to a basic operational problem. Advanced AI tools may sit on top of systems that were not built for the same speed, integration or scale. When those older systems become the source of truth, disconnected notes and outdated imagery can remain decisive.
Human review remains unclear
Under current US targeting doctrine, military commanders decide whether to prioritize and strike a target. They must distinguish military from civilian objects. The doctrine also includes an optional process called target vetting, which checks the accuracy of the underlying intelligence.
One former senior intelligence official told the Los Angeles Times it would be unthinkable for a commander to skip that step during strikes on the first day of a new campaign. Centcom reviewed targets before operations against Iran, but the source says it remains unclear whether the optional vetting process was initiated.
That uncertainty is important because AI targeting does not remove the need for verification. The system may help identify candidate targets, but the final decision still relies on confidence that the intelligence is current and that civilian objects are not being treated as military ones.
The Defense Intelligence Agency oversees both MIDB and MARS. When contacted by Bloomberg, it did not directly address the reported flaws or the delayed transition. A spokesperson pointed broadly to the thorough analysis conducted by assigned analysts.
The warning from an AI architect
Some targeting experts cited by the Los Angeles Times hope that better-connected digital systems and more AI could reduce errors. One possible improvement would be an automated cross-check against public services like Google Maps, which could flag anomalies for human review.
The Pentagon moved in that direction after the report by unveiling an agentic AI initiative. The logic is clear from the source: if systems can compare records, detect inconsistencies and surface warnings before a strike, they may help prevent old data from staying hidden.
But the sharpest criticism came from Jack Shanahan, a retired Air Force three-star general who was the first director of the Joint Artificial Intelligence Center established in 2018. Before that, he led Project Maven, the AI program tied to the broader shift toward military AI adoption.
Shanahan told the Los Angeles Times there is no excuse for a command failing to verify the accuracy of its intelligence. He also described targeting as a career field that had weakened while the military focused on counterterrorism. As early as 2017, he said, he could barely find people to fill those roles.
That assessment makes the school strike a warning about more than software. AI can help sort information, connect systems and flag mismatches, but it cannot compensate for every failure in data governance, institutional attention or human verification. The source shows that the most advanced part of the pipeline may be only as reliable as the older records and review practices underneath it.