Some AI news is exciting, some is odd, and then there are stories that stop you in your tracks. This week, a Bloomberg investigation into a Pentagon review gave us the hardest kind of AI story there is — what happens when overreliance on AI in a military targeting process meets data that was simply wrong. On February 28, during the opening days of the US–Iran war, two Tomahawk missiles struck the Shajarah Tayyebeh school in the southern Iranian town of Minab. More than 150 people were killed, including at least 123 children. The Pentagon's own investigators have now pieced together how it happened, and AI is central to the answer.
01.What the Pentagon investigation found
According to Bloomberg, which spoke with officials involved in the internal review, three things combined to produce the disaster. First, the administration had demanded an overwhelming aerial assault — more than 1,000 Iranian targets were struck within the first 24 hours — and that pace compressed the time available to verify each target. Second, the information about the site itself was outdated. The location had once been part of a military compound, but years earlier it had been physically separated by walls and had operated openly as a school. The satellite imagery showed the changes. The targeting database did not. Third, cuts to civilian-protection personnel meant that nobody on the remaining teams reviewed the strike plans before the missiles flew.
Outdated intelligence. Rushed timelines. Missing human checks. Any one of those alone is dangerous. Together, they were fatal.
02.Where Palantir's Maven comes in
Maven Smart System, built by Palantir, is an AI-enabled platform that pulls together intelligence from many data sources to help commanders identify potential targets and make decisions. It has become, in Bloomberg's words, a cornerstone of the US military apparatus over the past year. The problem, investigators found, was not that Maven malfunctioned. It's that some US Central Command personnel came to rely on it so heavily that they expected it to flag stale or inconsistent intelligence — a job the system was never built or trained to do. The AI was working exactly as designed. The design assumed the data flowing into it was current and correct. It wasn't.
Palantir, for its part, says it is "not responsible for the underlying data nor identifying intelligence deficiencies," and that there's no evidence its software was at fault. According to the reporting, the company has since built new tools to re-review underlying intelligence and flag inconsistencies that human review might miss.
03.The warning that never arrived
Here's the detail that honestly bothered me the most. A US intelligence analyst had reportedly noticed changes to the site as far back as 2019. But that warning was entered into a system that was not connected to the authoritative targeting database. So the information existed, inside the same government, and still never reached the people preparing the strike. This wasn't an AI failure in the narrow sense — it was an information-flow failure that an overworked, AI-accelerated process never caught. The Pentagon reportedly knew within hours that US missiles were responsible.
There's also a broader legal shadow over all of this. A UN-backed fact-finding mission said on Thursday that there are reasonable grounds to believe the US committed war crimes in two February airstrikes in Iran, one of them the Minab school strike. That's an allegation from a UN body, not a court verdict — but it guarantees this story is far from over.
04.The uncomfortable lesson for everyone who uses AI
Here's the thing: most of us will never touch a military targeting system. But the failure pattern here is universal, and every company in India racing to adopt AI should study it. US defence AI oversight is itself being rebuilt around the same time — the new 'AI Force' czar plan we covered is meant to be part of the answer. An AI system does not know when its inputs are stale. It does not know that the building on the map changed purpose years ago. It will confidently process outdated information and present the output with the same polish as accurate information — which is precisely when humans stop checking. Speed without verification isn't efficiency. It's risk, scaled up.
That's why the boring parts of AI — data hygiene, human review checkpoints, knowing what a system can and cannot do — deserve as much attention as the flashy parts. If you're learning to build with AI, make that a habit from day one. Our GPT-6 Astra — Zero to Hero Guide spends a full chapter on exactly this: where AI helps, where it fails quietly, and how to keep humans in the loop for decisions that matter. It's the chapter most people skip, and the one they shouldn't.
Key Takeaways
- Pentagon investigators found flawed intelligence, dated imagery and overreliance on AI contributed to the Minab school strike that killed 123 children
- Personnel expected Palantir's Maven to catch stale intelligence — a task outside the system's design; Palantir denies responsibility for underlying data
- A 2019 analyst warning about the site never reached the targeting database it needed to
- A UN-backed mission says there are reasonable grounds to believe war crimes occurred in two February strikes — including Minab
- The civilian lesson: AI amplifies whatever data it's given, so verification and human oversight are non-negotiable
Frequently Asked Questions
What happened in the Minab school strike?
On February 28, 2026, during the opening days of the US–Iran war, two Tomahawk missiles struck the Shajarah Tayyebeh school in the southern Iranian town of Minab. More than 150 people were killed, including at least 123 children, according to a Pentagon review reported by Bloomberg.
What role did Palantir's Maven Smart System play?
Maven is Palantir's AI-enabled platform that pulls together intelligence from many data sources to help commanders identify potential targets. Investigators found it did not malfunction — the problem was that US Central Command personnel relied on it so heavily that they expected it to flag stale or inconsistent intelligence, a job the system was never built to do. Palantir says it is not responsible for the underlying data and has since built new tools to re-review intelligence and flag inconsistencies.
Why didn't the 2019 warning prevent the strike?
A US intelligence analyst had reportedly noticed changes to the site as far back as 2019 — the compound had been physically separated by walls and operated openly as a school. But that warning was entered into a system that was not connected to the authoritative targeting database, so the information never reached the people preparing the strike. It was an information-flow failure that a rushed, AI-accelerated targeting process never caught.
What This Means for You
You might be a student, a developer, or a founder in India — and this story is still about you. Every AI system you'll ever build or use has the same skeleton: data in, decisions out. The teams that win long-term aren't the ones that automate fastest; they're the ones that keep asking, "what if the input is wrong?" Build that reflex now, while your projects are still small enough to learn from. And if you want a structured way to develop it — along with practical skills in prompting, agents, cost optimization and safe AI use — have a look at the GPT-6 Astra — Zero to Hero Guide. It's the cheapest insurance you'll find for an AI career.
By the way, if terms like agents and computer use still feel fuzzy, our GPT-6 Astra guide for beginners explains them all without the jargon.