A U.S. military investigation has revealed that artificial intelligence systems, combined with human errors and reduced civilian-harm oversight, played a significant role in a February strike on a school in Iran's Minab that killed more than 150 people, including at least 123 children. The findings expose critical vulnerabilities in how the Pentagon's AI-enabled targeting process, known as the "kill chain," handles intelligence verification and civilian protection.

What Went Wrong in the Minab Strike?

The investigation identified a cascade of failures that allowed outdated information to flow through the military's targeting system unchecked. The Minab site had been classified in the U.S. military's main intelligence database as an Islamic Revolutionary Guard Corps (IRGC) facility, but the location had changed significantly over the years and was operating as a school. Commercial satellite imagery showed that walls and entrances had been built to separate the school from the military compound, with a soccer pitch and other visible signs of civilian use.

One U.S. intelligence analyst had identified changes at the site as early as 2019, but those observations were recorded in a system that was not connected to the primary database used for targeting. This critical disconnect meant that decision-makers never saw the evidence that contradicted the outdated classification.

The timing of the strike compounded these failures. The U.S. military was working under intense time pressure before the February 28 assault, with more than 1,000 Iranian targets struck during the first 24 hours. Work that would normally take much longer to assess and approve potential targets was compressed into days or even minutes, according to officials familiar with the probe.

How Did AI Systems Contribute to the Targeting Error?

Maven, an AI system developed by Palantir Technologies, has become central to U.S. military operations. The system combines information from more than 150 data sources to help personnel assess potential targets and coordinate operations. However, the investigation revealed a critical assumption that undermined its effectiveness: some personnel appeared to expect the system to identify inconsistencies or stale information in the intelligence supplied to it.

The Minab site passed through multiple stages of the military's targeting process with high confidence that it was an IRGC facility. One Pentagon official involved in the probe remarked on the false certainty that preceded the strike, stating:

"Minab proved you should never be too confident."

Pentagon official involved in the investigation

Palantir has rejected responsibility for the intelligence failures, arguing that the company "is not responsible for the underlying data nor identifying intelligence deficiencies" and that there was no evidence its software was at fault in the Minab strike.

What Systemic Failures Made the Strike Possible?

Beyond AI limitations, the investigation uncovered severe staffing and procedural breakdowns in the Pentagon's civilian-harm assessment process. The targeting process, known as the "kill chain," involves multiple stages including intelligence analysis, imagery review, weapons planning, legal assessment, and commander approval. However, staffing across civilian-harm assessment units had been cut sharply, with the Central Command team reduced from 10 members to one.

No member of the civilian-harm assessment team reviewed the Minab site before the strike. Such reviews had previously been used to assess civilian activity around potential targets and identify ways to reduce the risk to noncombatants.

The key failures in the targeting process included:

• Outdated Satellite Imagery: The military relied on satellite data that did not reflect the site's current use as a school, missing visible evidence of civilian occupation.
• Disconnected Intelligence Systems: Critical observations about changes at the site were stored in a database separate from the main targeting system, preventing decision-makers from accessing contradictory evidence.
• Reduced Civilian-Harm Staffing: Civilian protection teams were severely understaffed, with the Central Command unit reduced to a single person who did not review the Minab target.
• Time Pressure and Compressed Workflows: The accelerated pace of targeting during the 24-hour assault compressed multi-day review processes into minutes, reducing opportunities for verification.
• Over-Reliance on AI Without Verification: Personnel expected Maven to flag inconsistencies in intelligence, but the system was only as good as the data fed into it.

How Has the Pentagon Responded to These Failures?

Since the attack, U.S. Central Command has implemented significant changes to its targeting process. The modifications include additional procedures to refresh and vet planned targets, new open-source data feeds to provide more information about civilian activity, and dozens of upgrades to Maven.

Palantir has also added capabilities that allow the system to re-examine underlying intelligence and flag inconsistencies or inaccuracies that may have been missed during human review, according to people familiar with the changes. Cameron Stanley, head of the Pentagon's digital and artificial intelligence office, noted that the upgrades reflected lessons from the conflict, stating:

"Different types of things were identified, different challenges were identified, new data sources were incorporated."

Cameron Stanley, head of the Pentagon's digital and artificial intelligence office

However, these changes come after the fact. The investigation demonstrates that AI systems alone cannot replace human judgment and institutional safeguards, particularly when those safeguards have been weakened by budget cuts and time pressure.

What Are the Broader Implications for Military AI?

The Minab investigation raises fundamental questions about how the U.S. military deploys AI in high-stakes targeting decisions. The findings suggest that AI systems can amplify human errors rather than correct them, especially when personnel misunderstand the system's capabilities or when critical institutional checks have been removed.

The United Nations has separately raised concerns about the strike. A UN fact-finding mission said this week that there were reasonable grounds to conclude that the Minab attack and another U.S. strike amounted to war crimes; a finding the U.S. has declined to acknowledge.

As defense AI systems like Maven become more central to military operations, the Minab case illustrates the risks of treating automation as a substitute for institutional oversight. The investigation's findings suggest that future AI-enabled targeting systems must be paired with robust verification procedures, adequate staffing for civilian-harm assessment, and realistic timelines for decision-making, rather than relying on AI to compensate for human and institutional failures.