Guardian: Minab School Bombing Wasn't a Claude Failure

Kevin T. Baker argues in the Guardian that Anthropic's Claude has been miscast as the villain of the February 28 strike on the Shajareh Tayyebeh girls' school in Minab, and that the real story sits inside a targeting bureaucracy engineered to run faster than any human could check it.

The school building sat on a Defense Intelligence Agency target list for years, classified as a military facility long after it had been walled off and converted to civilian use. It appeared in Iranian business listings and was visible on Google Maps. A search engine could have found it. Nobody searched.

"The building was on a target list for years. Yet this was missed, and the question is how," a former senior government official told Baker.

Baker's point is that the speed itself made the miss inevitable. Palantir's Maven Smart System, the platform running the strikes, is built to sustain roughly a thousand targeting decisions an hour, about 3.6 seconds each. Claude and other large language models were added late in 2024 as a natural-language layer over Maven's sensor-fusion engine, not as the thing choosing what to hit. "LLMs-gone-rogue dominated coverage, but had nothing to do with the targeting," Baker writes. "Instead, it was choices made by human beings, over many years, that gave us this atrocity."

Two analysts we track in our Who's Who directory were circulating the essay within a day of it running.

The piece does not publish any internal DIA review, and the criminal-accountability question stays open. What Baker supplies instead is a bureaucratic history stretching from Project Maven's 2017 origins through Palantir's LLM integration, and a plain warning about a kill chain squeezed to 3.6 seconds: the verification steps that used to catch a stale database entry are the ones cut first.