The first civilian confirmed killed in an AI-assisted strike?

On 10 March 2026, a joint investigation by Airwars and The Independent appeared to identify the first time the U.S. military has ever officially accepted killing a civilian in an AI-assisted strike. The Independent articles, written by Namir Shabibi of Airwars and Alex Croft of The Independent, can be read here and here. Airwars’ version is below.

The U.S. military has “no way of knowing” whether it used artificial intelligence in conducting specific airstrikes, a spokesperson has claimed.

The statement came in response to a joint investigation by Airwars and The Independent that appeared to have identified the first time the U.S. has officially accepted killing a civilian in an AI-assisted strike.

20-year-old student Abdul-Rahman al-Rawi was killed in a U.S. airstrike in Iraq in February 2024. At the time a top U.S. official said the strikes used AI targeting and the military later accepted al-Rawi was killed unintentionally – even sending his family a letter of condolence.

But the military is now claiming it cannot know whether the strike that led to his death was AI assisted – with a spokesperson saying they cannot tell whether specific strikes relied on machine learning algorithms.

The revelations come amid intense debate around AI in warfare, with reports that AI was used in both U.S.-Israeli strikes on Iran, and by the U.S. military in January’s raid in Venezuela. In a major standoff with the Pentagon, Claude AI’s maker Anthropic recently announced its refusal to remove safeguards preventing use in surveillance and fully autonomous weapons.

‘No recognition, just misery’ 

On the night of 2-3 February 2024, the U.S. launched strikes against more than 85 targets, comprising Iraqi-government aligned forces and Iranian-backed militias in Iraq and Syria. It followed a drone attack on a U.S. base in Jordan which killed three military contractors.

U.S. Central Command (CENTCOM) stated that targets included command and control operations centres, rocket launchers, and supply chain facilities. The Iraqi government condemned the strikes – saying that among those hit were “Iraqi security forces as well as civilian sites.”

That month CENTCOM’s chief technology officer, Schuyler Moore, declared the targets had been identified with support from machine learning algorithms known as Project Maven. It was the first such public declaration about individual strikes from the U.S. military.

“We’ve been using computer vision to identify where there might be threats,” Moore told Bloomberg.

Up to three civilians were reported to have been killed and in a review quietly released last summer, the Department of Defense (now Department of War) admitted one civilian casualty in the strikes – saying that the family of the victim had been contacted.

Airwars identified that victim as Abdul-Rahman al-Rawi, a 20-year-old college student studying construction.

The U.S. military’s admission would mark the first time a military has ever acknowledged civilian harm arising from AI-assisted strikes. Other nations have used AI technologies to identify targets for strikes – most prominently Israel during the war in Gaza – but have never officially accepted killing civilians in those strikes.

An image of the car that was hit in the strike. Abdel Rahman Al-Rawi was standing only metres away, his family said. Picture provided by his family.

Abdul-Rahman was killed after a strike on a parked car just meters away from him, his brother Anmar al-Rawi said in an interview with Airwars.

“Civil Defence arrived at the scene to put out the fire. They searched around the vehicle with flashlights and found human remains. I had to identify them as that of my brother.”

The proximity of the blast to Abdul-Rahman tore his body to pieces… It took us two days to gather all of my brother’s remains”.

“My dad remains deeply depressed. My mother suffered a heart attack and high blood pressure ever since. Every time she is reminded of her baby boy she breaks down,” Anmar, a nurse in the local hospital, explained.

An image of Abdul-Rahman al-Rawi, provided by his brother

Asked whether AI was the source of the error in the strike, a CENTCOM spokesperson for Operation Inherent Resolve (OIR) told Airwars and The Independent that “Following the strike on 2 Feb 2024, a comprehensive internal investigation was conducted. There is no indication that AI was used at any point in the targeting process.”

Airwars has reviewed hundreds of U.S. military civilian harm assessments, including the more than 1,000 released to
The New York Times in 2022. Their primary purpose is to assess reports that civilians were harmed, and have not to date explored or investigated the technology underpinning the targeting selection. CENTCOM did not elaborate further.

Apparently contradicting their earlier statement, the CENTCOM spokesperson went on to state that it could not determine whether AI was used in the strike that killed Abdul-Rahman.

“We have no way of knowing whether this strike is one of the 85 [strikes] that Ms. Moore described [as having been targeted using AI],” the statement said.

Airwars obtained a copy of a letter from the Department of Defense addressed to Al-Rawi’s elder brother, dated 29 January 2025, in which it expressed “condolences…[and] our deepest sympathies” over al-Rawi’s death.

The letter, signed by the then U.S. Air Force Chief of Staff, made no mention of the circumstances of the strike, or what accountability mechanisms would follow.

The letter of condolence sent to Abdul-Rahman’s family. Provided to Airwars by the family

While the Iraqi Red Crescent offered 250,000 Iraqi Dinars (or 141 GBP) to assist with repairs to damage sustained to the al-Rawi family home, Abdul-Rahman’s brother says they did not receive any form of compensation or ‘ex gratia’ payment from the U.S. military. The CENTCOM spokesperson declined to comment on the matter.

“There has to be some sort of recognition and apology. But we got nothing from them, just misery,” Anmar said.

Up to five medical personnel belonging to the Iraqi Popular Mobilisation Forces (PMF) were also reported killed, along with damage to a field hospital, reports claim. Medical workers and their facilities are deemed protected under international humanitarian law.

The PMF are Iranian-backed forces that played a key role in defeating Islamic State forces in Iraq, and were partially integrated into the state a decade ago.

AI-assisted targeting

Militaries are integrating AI into all aspects of the so-called ‘kill chain’ in a bid to speed up targeting and gain advantages over rivals.

Initially developed by the Pentagon, Project Maven was adopted by the National Geospatial Agency (NGA), and uses computer vision algorithms to locate and identify targets from satellite imagery, video and radar to detect movement and track targets.

Project Maven saw its first major deployment following Russia’s invasion of Ukraine in 2022, albeit with a “basic” version provided to Ukrainian forces to help identify Russian military vehicles, people and buildings.

However, Maven delivered mixed results in the war that ensued. Snow, dense foliage and decoys are known to hinder its abilities. And in desert terrain like western Iraq, where weather conditions can change a landscape abruptly, Maven’s accuracy can drop to below 30%.

“At the end of the day this [Ukraine war] became our laboratory,” the commander of the 18th Airborne Division said at the time. Based at Fort Bragg, North Carolina, the 18th Airborne Division was behind the targeted strike in Iraq and Syria on the night of 2 February 2024.

Maven is now available to all U.S. services and combatant commands and, since the strikes in 2024, its user base has more than quadrupled, then-NGA Director Vice Adm. Whitworth said in a keynote last year. It is currently able to make 1,000 targeting decisions in an hour, “choosing and dismissing targets on the battlefield,” he explained.

A month later, Whitworth acknowledged that the NGA was using artificial intelligence so routinely that it created a standardised disclosure to go on AI-generated intel products. “We want to use it for everything, not just targeting,” Whitworth said.

Project Maven is typically integrated into Palantir’s Maven Smart System (MSS), an AI-enabled warfighting system, to speed up U.S. military targeting decisions. Palantir’s MSS, which uses Anthropic’s Claude AI, is currently deployed by the U.S. to assist targeting in Iran.

Whitworth previously stated that the small 20-person targeting team based at Fort Bragg had exceeded the targeting performance of the 2,000-strong cell behind the 2003 invasion of Iraq, through combined use of Project Maven and Palantir’s MSS.

“The benefit that you get from algorithms is speed”, CENTCOM’s chief technology officer told Bloomberg following the strikes on Iraq. However, with speed has come growing concerns about the ‘human in the loop’ being little more than a ‘rubber stamp’ for machine-made targeting recommendations.

Experts have warned that current frameworks fail to address the “profound risks” that ‘decision support systems’ like Project Maven pose to international humanitarian law and human judgment in targeting. These concerns have been echoed by tech workers opposed to their companies’ involvement in engineering AI systems for warfare.

Initially a key player in Project Maven, protests and resignations from Google employees against the company’s involvement in artificial intelligence for lethal purposes saw the company exit the project. Palantir stepped in to fill the void, referring to the project internally as ‘Tron’, after the 1982 film.

Revelations that Claude AI was used in the U.S. raid on Venezuela in January has led to ongoing tensions between its maker, Anthropic, and the Pentagon. Anthropic does not permit its AI systems to be deployed for mass domestic surveillance or autonomous lethal weapons.

In a punitive move, on Thursday the Pentagon designated Anthropic a “supply chain risk” with major consequences for the company. Despite this, reports suggest Claude is being widely utilised by the U.S. military in bombing Iran.

Israel’s offensive in Gaza has seen it deploy AI in a number of ways – a target-creation platform called “the Gospel” which used machine learning algorithms to produce potential targets so fast Israeli officers have compared it to a “mass assassination factory”.

Another AI-assisted target identification tool, called ‘Lavender’, at one stage identified 37,000 potential targets based on their apparent links to Hamas. One Israeli intelligence source downplayed the significance of human oversight in Lavender’s target selection: “I would invest 20 seconds for each target at this stage, and do dozens of them every day. I had zero added-value as a human, apart from being a stamp of approval.”

The U.S. military’s admission it ‘has no way of knowing” what role AI is playing in identifying targets for aerial strikes raises urgent questions amidst the current war on Iran.

U.S. Central Command (CENTCOM) has employed a variety of AI tools to provide real time targeting suggestions in a ferocious volley of attacks across Iran, which authorities say have killed more than 1,200 people. The scale of bombing is unprecedented: an analysis by Airwars last week revealed the first days of bombing in Iran saw far more sites targeted than any recent U.S. or Israeli military campaign. The scale of the campaign is likely to intensify scrutiny over accountability, transparency, and the expanding role of AI in warfare today.