The Pentagon is letting software push strike decisions at a pace humans cannot match. In Iran, the US military says it hit more than 2,000 targets in four days.
That tempo is being fed by AI systems that chew through intelligence from drones, satellites, and other sensors, spitting out strike options faster than old-school planning. The fighting has become the first battlefield outing for so-called frontier generative AI models, the sort civilians use at desks and in clinics.
Over the past two years, the US Department of Defence has integrated AI more deeply into day-to-day operations. The Pentagon’s core data layer runs on Palantir’s Maven Smart System, paired with Anthropic’s Claude to provide a real-time operational dashboard in Iran.
Palantir UK and Europe head Louis Mosley reckons the tech has shifted from text tidying to something closer to structured thinking. That change, he argues, is pushing up both the volume of decisions and the speed at which military personnel take them.
The rush comes with a nasty side effect: reduced oversight when targets are generated and acted on at machine speed. The debate has sharpened after a dispute between Anthropic and the Pentagon over where to draw the lines on military AI.
A bombing of a girls’ primary school in Minab in southern Iran shows how lethal a bad call can be, though it is unclear how much AI had a hand in it. Iran’s Red Crescent says more than 20,000 non-military buildings have been hit, including 17,353 residential properties.
Utrecht University researcher Jessica Dorsey put the scale in context by comparing it with the campaign against Isis. She noted that coalition forces struck about 2,000 targets in the first six months in Iraq and Syria, versus four days here.
In a live operation such as Operation Epic Fury, Palantir’s Maven platform is treated like the software brain behind the kill chain. It covers spotting and ranking targets, picking weapons, and assessing battle damage.
A defence tech expert, who asked not to be named, said: “Those [older] kill chains are measured in hours and sometimes days,” with commanders poring over printouts before signing off.
Cambridge University researcher Sophia Goodfriend said: “It supersedes anything we’ve seen in the past, and it allows militaries to operate at unprecedented speed and scale in aerial targeting.”
The user base behind Maven is not small either, with public comments saying it had more than 20,000 users across 35 military entities in the field by May 2025. Defence researchers think the figure in the US may be closer to 50,000 now, with NATO signing up in 2025.
AI is turning up across other parts of the war machine, from autonomous navigation to computer vision, and not just in Iran.
Oslo Nuclear Project missile expert Fabian Hoffmann said: “That was too difficult, but now with drones and AI, this is surely becoming easier,” as image recognition scans footage to spot launchers and other kit.
