
U.S. Army Maj. Steven McPherson, a joint doctrine team chief with the National Guard Bureau, reviews the Maven Smart System interface during training in Arlington, Va. (Whitney Hughes/U.S. Army)
The Pentagon is investing heavily in artificial intelligence to help speed up battlefield decisions, a push that many military ethics experts warn could leave commanders increasingly reliant on machines when deciding how to use lethal force.
Those concerns have grown during the U.S. war against Iran as American forces have carried out thousands of strikes while acknowledging the use of AI systems to help accelerate targeting.
Among the deadliest U.S. attacks was a strike on a school in Minab, Iran, on Feb. 28 that killed more than 150 people, most of them children, according to Iranian authorities. The strike has prompted questions from Democratic lawmakers about whether AI played a role in the targeting process.

A man hold a children’s backpack in the aftermath of an Israeli-U.S. strike on a girls’ elementary school in Minab, Iran, Saturday, Feb. 28, 2026. (Abbas Zakeri, Mehr News Agency via AP)
The risk of deadly targeting mistakes is not new. Such errors occurred long before artificial intelligence was able to assist in selecting targets.
The question is whether the growing use of AI could increase those risks by encouraging commanders to accept machine-generated recommendations without properly questioning them, according to some experts.
“There’s a human propensity to trust the machine,” said C. Anthony Pfaff, director of the research and assessments department at the Army War College. “We give in to that too easily.”
U.S. military leaders have defended the growing use of AI, saying it helps troops make sense of large amounts of battlefield information.
“Humans will always make final decisions on what to shoot and what not to shoot, and when to shoot,” said Adm. Brad Cooper, commander of U.S. Central Command, in a Defense Department video in March, just days after the Minab bombing. “But advanced AI tools can turn processes that used to take hours and sometimes even days into seconds.”
A Pentagon investigation into the Minab strike found that outdated intelligence contributed to the attack, according to September reporting by Bloomberg, which cited officials familiar with the inquiry.
The school had once been part of a military compound but had been separated from it years earlier. Bloomberg reported that an analyst identified changes to the site in 2019, but the information was stored in a system disconnected from the primary targeting database.

This satellite photo from Planet Labs PBC shows the area of a school and base of Iran’s paramilitary Revolutionary Guard after what appears to be an airstrike targeting the compound in Minab, Iran, Wednesday, March 4, 2024. (Planet Labs PBC via AP)
Investigators speaking to the news outlet also raised concerns that CENTCOM personnel relied too heavily on the Maven Smart System, expecting the AI-enabled platform to flag outdated or conflicting intelligence.
Palantir Technologies, which developed Maven, disputed suggestions that its software contributed to the strike, saying it was not responsible for the underlying intelligence and that there was no evidence its technology caused the error.
Following the attack, CENTCOM reportedly revised its targeting procedures and upgraded Maven to help identify inconsistencies in intelligence, according to Bloomberg.
The Pentagon has not publicly released the results of its investigation into the Minab strike. CENTCOM did not respond to a Stars and Stripes request for comment on the investigation’s status or Bloomberg’s reporting that the command revised its targeting procedures and upgraded Maven Smart System following the attack.
Questions about the role of AI in targeting have surfaced in investigations of previous American military operations.
An investigation earlier this year by Airwars and The Independent examined the death of Abdul-Rahman al-Rawi, a 20-year-old college student killed during U.S. airstrikes in Iraq in 2024.
A senior U.S. military official had previously said machine-learning technology helped identify targets during the broader operation. But CENTCOM also told investigators that an internal review found no indication AI was used in the strike.
The emerging concern is whether AI-assisted systems could amplify existing risks by accelerating decisions or making errors more difficult to detect.
“Everybody can do everything right, and something can still go wrong,” Pfaff said.
A U.S. Marine controls a Bolt-M loitering munition during a range event at Camp Lejeune, N.C., Aug. 28, 2026. (Jorge Borjas/U.S. Marine Corps)
And distributing decision-making among multiple people and technological systems can also make determining responsibility difficult in those situations, he added.
That risk is particularly worrisome when it comes to the process of identifying, assessing and engaging targets, known in military parlance as the kill chain.
Anthony Randall, a retired Army chaplain who has spoken publicly about the ethical implications of AI in combat, said the push for speed could reduce opportunities for self-reflection.
“As the kill chain accelerates because of AI, where are there still those strategic nodes of human perception and human reason that impact our moral, ethical and spiritual decision-making?” Randall said.
For Randall, a further fear stems from how a service member’s psyche could suffer as a result.
Relying on AI-assisted recommendations that do unintended harm, even when established procedures were followed, might be deeply distressing and adversely affect troops’ well-being, he said.
The challenge for commanders is using AI to make faster decisions without relying too heavily on it when lives are at stake.
Pfaff said that responsibly designed AI systems may reduce civilian casualties and targeting errors by helping military personnel evaluate information more thoroughly.
But even when AI improves battlefield decisions, commanders must be careful not to surrender their own judgment to machines, he said.