TL;DR
AI-powered decision support systems are being used in military operations against Iran, pulling together satellite imagery, intercepted communications and social media streams to flag targets and rank threats. Experts warn that the speed advantage these systems provide may be eroding meaningful human oversight into “rubber stamping.”
How AI Is Being Used
The systems in question — called “decision support systems” — don’t fire weapons. They process thousands of data inputs from satellite imagery, intercepted communications, logistics data and social media, surfacing patterns far faster than any human team. The human still presses the button.
Israel has used similar AI systems in Gaza to flag potential targets. The US military reportedly used Anthropic’s Claude model during its operation in Venezuela and continued using it during strikes on Iran, even after Anthropic’s fallout with the Trump administration.
Defence Secretary Pete Hegseth sent a memo early this year directing the Department of Defense to “accelerate America’s Military AI Dominance by becoming an ‘AI-first’ warfighting force across all components.”
The Oversight Problem
The central reassurance in military AI has always been “a human in the loop.” OpenAI reinforced this after announcing its Pentagon partnership, stating it had secured agreement against three red lines: mass domestic surveillance, autonomous weapons, and high-stakes automated decisions.
But researchers question whether loop-based oversight works in practice during active conflict.
“Humans are technically in the loop,” says Craig Jones, senior lecturer in political geography at Newcastle University. “That doesn’t mean they are in the loop enough to have effective decision-making power.”
David Leslie, professor of ethics at Queen Mary University of London, puts it more bluntly: “We are really facing a potential scaled hazard of rubber stamping, where because of the speed involved, you don’t have active human, critical human engagement.”
AI’s Own Limitations
Beyond oversight, there is the question of reliability. AI systems guess what is most probable based on training data — and most of the time, that statistical reasoning works. But warfare is unpredictable by nature.
Research has shown AI models confidently producing wrong answers and insisting they are correct, raising concerns about deploying these systems in environments where errors carry the highest possible stakes.
Looking Forward
There appears to be no reversal on the horizon. The US military’s directive is clear: adopt AI quickly and at scale. The question now is whether governance frameworks can keep pace with deployment speed — particularly for the UK, which signed a £240m contract with Palantir for AI-powered defence analytics last year.