May 2026
Written by Michael Clare
In the US air and missile campaign against Iran, Operation Epic Fury, which began on February 28, artificial intelligence played a major role in selecting attack targets. The Pentagon is using the Maven Smart System, an AI-powered data fusion and decision support program, to identify high-priority targets and help select weapons for attack, senior U.S. officials said.
According to an April 8 White House report, in the first 38 days of the war, U.S. forces attacked more than 13,000 targets inside Iran, including more than 2,000 command and control targets, more than 1,500 air defense targets, and more than 1,450 industrial base targets. The White House insisted all were legitimate military targets, but new york times and other news media reported visual destruction of civilian facilities. This has led critics to argue that over-reliance on AI has led to targeting errors and unnecessary civilian casualties.
The Maven Smart System, manufactured by Palantir Technologies, as employed by Iran, collects information about enemy locations from radar signals, satellite and drone imagery, electronic communications and other sources and integrates it into a “common operational picture” of the battlefield, Cameron Stanley, the Pentagon’s chief digital and AI officer, said March 12 at Palantir’s AIPCon 9 conference. He said Maven is also used to attack specific targets and identify friendly forces in the best position to conduct attacks. The course of action that U.S. strike forces should follow, including attack vectors, munitions used, and other relevant data.
Instead of “eight or nine systems” that commanders refer to when making combat decisions, Maven can fuse them all into a “single visualization tool” that can be used for decision-making, Stanley explained. Equipped in this way, commanders can identify targets and select attack packages to attack them with a simple click of the screen, he said. “So we could identify a target, develop a course of action, and act on that target all from one system. This is revolutionary.”
Stanley said Maven has become increasingly sophisticated over time. Launched in 2017 by the Pentagon’s Algorithmic Warfare Multifunctional Team (later the Joint Artificial Intelligence Center), Project Maven was designed to use computer vision and machine learning to classify drone footage of battlefields in the Middle East to identify extremist hideouts that could lead to attacks. Since then, Palantir has steadily improved its technology, now renamed the Maven Smart System, which can collect and collate data from multiple sources and recommend possible combat moves.
In late 2024, Anthropic’s Claude AI operating system was integrated with Maven technology to provide military users with enhanced targeting options. Using an integrated system, commanders can now generate lists of targets based on a variety of criteria, such as radar stations, missile emplacements, communications nodes, and senior commanders, and rank them by strategic importance. When a target is attacked, the system reviews the damage assessment report and automatically creates a new target list. All this takes just a few minutes.
Anthropic is currently prohibited from providing services to the U.S. military due to its refusal to work on autonomous weapons systems or domestic surveillance operations. Other AI companies, including OpenAI, were also hired to fill Anthropic’s role.
To U.S. military officials, Maybin’s speed of target selection and reselection represents a clear combat advantage, allowing U.S. forces to quickly and relentlessly disable Iran’s combat capabilities and prevent Iranian reconfiguration.
“Our warfighters utilize a variety of advanced AI tools,” Adm. Brad Cooper, commander of U.S. Central Command, said in a March 11 video conference. “These systems help sift through vast amounts of data in seconds, so leaders can cut through the noise and make smart decisions faster than their adversaries can react,” Cooper said.
However, it is this very speed of target selection that has many observers concerned, given the risk that sites for attack will be selected without proper human oversight. Officials are said to scrutinize every target before a strike order is issued, but there is a growing risk that people will become “too reliant on the system” and not reaffirm its recommendations, Nirza Amaral, research director at Chatham House’s Center for Global Governance and Security, told Abu Dhabi-based newspaper The National.
Many analysts say the risk of error is naturally increasing because humans have less and less time to review AI targeting decisions. It is unclear whether AI played a role in the February 28 US cruise missile attack on Shajare Tayebeh Girls’ Elementary School in Minab, Iran, which resulted in the deaths of more than 170 people, mostly children. According to a preliminary assessment by Central Command, U.S. intelligence maps do not show that the school facility, which was once part of a military base, was long ago converted to civilian use and added to the AI-generated target list without proper human oversight. new york times “Officials said the error was likely due to wartime human error rather than new technology,” the March 11 report said.
While the most likely cause of the Shadjare-Tayebeh missile attack is thought to be human error, a failure to update military intelligence maps, many observers warn that increased reliance on AI-powered targeting systems, a phenomenon known as “automation bias,” will cause more tragedies of this kind.
“Targeting is a concern.” [approval] “Automation bias can make it a mere formality, with people simply relying on the machine’s instructions,” Amaral said.
