Recent reports have spotlighted growing concerns about Israel’s purported use of artificial intelligence in fixing bombing targets during strikes on Iran, critics arguing that a lack of human review has led to mistakes such as hitting civilians. The systems were first tested in Gaza, analysts say, and are now being used against Iranian sites — a potential harbinger of an era of autonomous “killer robots.” This development has released an international debate on AI in warfare ethics and regulations.
Background on AI Targeting Systems
During the war in Gaza, Israel also developed AI tools it imported from other branches of military to select targets with surgical precision: It could approve 20-second strikes that would tolerate a 10 percent error rate. These systems waffle through vast troves of intelligence data to classify threats, focusing quantity over quality, according to insiders. The same technology allegedly fueled recent operations against Iran’s nuclear program and the Revolutionary Guard, joining an AI analysis with Mossad spies and smuggled drones.
Key Accusations from Experts
Trita Parsi of the Quincy Institute, accused Israel of bombing Tehran’s “Police Park,” a civilian location successfully misidentified by AI as government-related with no verification, in its foreign relations the same way a military attack into Gaza pattern. Tech journalist Jacob Ward has called this “lethal beta,” with Gaza as the prototype and Iran a full launch, supercharging an M.A.D. AI arms export industry. Critics say policies that permit up to 20 civilian deaths per strike deepen AI’s faults, yet provide no evidence they reduce harm.
Operation Details and Casualties
Israel applied the technology in an operation called Operation Rising Lion, executing a plan — with genuflection to President Trump — after having scoured data everywhere for records of any generals’ movements and striking targets like Gen. Hossein Salami and nuclear scientists. Over 1,000 Iranians were shot by strikes, including 175 children in what they dubbed a Minab girls’ school “double-tap.” US systems such as Palantir’s Maven and Anthropic’s Claude helped choose 1,000 targets in a span of 24 hours, but Trump ordered to opt-out
| Aspect | Gaza Precedent | Iran Application |
| AI Role | Habsora for mass targeting | Data sifting for nukes, generals |
| Oversight | Mid-rank approvals, 20 civ risk | Alleged none; park bombed unchecked |
| Casualties | 126 in Jabalia (68 kids) | 1,000+ total, 175 kids |
| Tech Ties | Project Nimbus (Google/AWS) | US Maven, smuggled drones |
Broader Implications and Calls for Regulation
Experts like Paul Scharre say that “AI gets it wrong” in life-or-death situations and need for human checks. UN negotiations in Geneva are about prohibiting completely autonomous weapons because people are worried about them spreading. Supporters say it will be faster, but researchers like Craig Jones don’t find any evidence that it will lead to fewer civilian casualties and may might lead to more.