Israel’s AI versus the ayatollah

In just 11 days, Israel and Iran engaged in what might be remembered as the world’s first true Artificial Intelligence (AI)-enhanced state-on-state war. But only one side showed up, Israel. The ayatollah brought a Structured Query Language query to a Python fight. Iran was fighting like it’s still 2022.

This wasn’t a conflict defined by territory or troop movements. It wasn’t even about nuclear facilities or missile interceptions. It was about tempo. Israel didn’t just move faster, it thought faster. This war showed clear AI asymmetry: Israel didn’t just dominate Iran’s sovereign airspace, it dominated its cognitive space. Israel’s algorithms were writing death sentences faster than Iran’s generals could read their morning briefings.

At the heart of that dominance: Artificial Intelligence fused with data intelligence, human intelligence, and tactical cunning. AI didn’t replace soldiers or spies. It augmented them. It extended their reach, amplified their speed, and stripped the friction out of kill chains. That’s what AI can do.

But it can’t do everything.

Israel, for all its technological supremacy, absorbed significant damage, and the ayatollah remains unbowed. Israel’s rocket interceptor stockpiles are running low. The new Jewish Space Laser isn’t yet battle ready. Ballistic missiles got through, causing death, destruction, and displacement. A tiny country 80 times smaller than Iran and already at war on seven fronts, Israel cannot sustain a protracted war of attrition, especially one more than 2 000km away. As for the stated aim of this war, no-one knows with credible certainty if an Iranian atomic bomb is now more or less likely.

This war will be benchmarked

The opening shots of the war were all about tempo and precision. At 04:30 on Friday, 13 June, without warning and without visible operators, Mossad’s pre-assembled missiles came alive. Iranian air defence systems, and the sense of security that came with them, were erased in seconds. The strike reportedly used commercial-grade components like Moxa ioMirror E3210 controllers – hardware you could buy off eBay – combined with thermal cameras and pre-programmed self-destruct routines. Once used, the communications boxes were obliterated. The cameras, conspicuously intact. Message received: we see you, and we leave nothing behind but your confusion.

This was IKEA + AI warfighting: pre-assembled, sleek, modular, self-guided, and disposable.

Next, Iran’s top military brass and nuclear scientists were assassinated, many within the first 15 minutes. Most of them were taken out at home or in their offices. Next came a wave of precision bombings against the country’s air defences, rocket launchers, rockets, and key units. Within a few hours, Israel delivered the kinds of blows previous armies took years to achieve, if ever.

This isn’t simply about military capacity, it’s about the speed of perception and the paralysis of the adversary. While Iran was still counting how many generals it had lost, Israeli analysts were training the next model on new data on the generals’ replacements.

Every Iranian move became training data. Israel’s AI was in a reinforcement learning loop. Each Iranian response and movement taught the algorithm something new about Iranian capabilities, decision-making patterns, and vulnerabilities.

Israel was fighting a war to dominate the other side’s OODA loop (observe, orient, decide, act), and turning it into a loading screen. And while Iran’s Revolutionary Guards still believes in flags, parades, martyrs, and sacred fire, Israeli forces swear by latency, inference, and proprietary chipsets. Kill chains are now written in Python, not Persian.

Israel’s advantage wasn’t just algorithmic, it was architectural. While Iran’s command structure still relies on human-in-the-loop decision trees, Israeli systems have moved to human-on-the-loop automation. The difference: milliseconds versus minutes; inference versus deliberation. Israel sees the battlefield as a training dataset where engineers push updates in real time. The result was clear to see.

But here’s where things get messy and AI isn’t able to flip the script. While the cost of attacking is collapsing, the cost of defending is skyrocketing.

Iran reportedly fired off about 20% of its ballistic missile stockpile, yet its attacks were relentless, forcing millions of Israelis into bomb shelters day after day, night after night. It was nerve-wracking, exhausting, and damaging. And though Israel and the United States intercepted about 90% of the incoming rockets, the hit rate was dropping. Israel’s missile defence systems – the Arrow, David’s Sling, and Iron Dome – aren’t cheap. And they were running hot. You don’t need a degree in economics – or warfare – to see the problem.

Despite its abilities, AI’s blind spots are glaring. Algorithms excel at pattern recognition but struggle with strategic surprise. Iran’s asymmetric responses – proxy warfare, cyber-attacks, and diplomatic pressure – don’t fit neat training datasets. AI can predict where a missile will land, but not why a leader might suddenly sue for peace or double down on destruction.

AI can win battles, but it can’t win the peace we crave. Israel has achieved something extraordinary: real-time kill chains; pre-emptive targeting; multi-domain tempo control. But turning that into a strategic endgame is a human political problem. Israel just executed the first modern war of inference. But even with the fastest systems, peace remains a decidedly analogue affair.

This war will be studied, not just by military tacticians, but by AI ethicists, technologists, and policymakers for years to come.

• Award-winning writer, editor, and host of The Dejargonizer podcast, Amir Mizroch was born in Israel and raised in Krugersdorp, South Africa, studying journalism at Rhodes University. He worked at the Mail & Guardian before making aliya in 2000. In Israel, he worked at The Jerusalem Post and Israel Hayom, and was later director of communications at Startup Nation Central. See his Substack: Israeltechinsider.com