AI Targeting in War: Assessing Risks to Civilians under International Humanitarian Law

Recent armed conflicts increasingly rely on artificial intelligence (AI) technologies to assist or inform military decision-making, particularly in target identification, surveillance, and strike selection. For instance, AI-based targeting has played a significant role in recent conflicts, particularly in Gaza, Ukraine and Iran. Although AI technologies are subject to certain ethical principles, these are not always properly followed. While such technologies are often promoted as improving precision and reducing collateral damage, their use has raised serious legal and humanitarian concerns regarding civilian protection, because their development and deployment are not always sufficiently human-centred.
In the conflict in Gaza, the Israeli Defence Forces (IDF) reportedly used an AI system known as “Lavender,” which had an estimated 10 per cent error rate that created life risks to civilians. By March 2024, using “Habsora,” which selects bombing targets, more than 30,500 people had been killed in Gaza by the IDF. In the war between Russia and Ukraine, both sides used AI-enabled drones widely for attacks. According to UN-verified data from the UN Human Rights Monitoring Mission in Ukraine (HRMMU), more than 15,172 civilians had been documented as killed since the full-scale invasion began on February 24, 2022, while 41,378 were seriously injured. At least 766 children were among those killed. The widespread use of AI-enabled weapons and decision-support systems in these conflicts has raised serious concerns about civilian protection, accountability and compliance with international humanitarian law.

Within the first six weeks after October 7, 2023, Lavender generated around 37,000 target recommendations with a 10 per cent error rate, which means thousands of civilians may have been misidentified and killed indiscriminately.

In the recent USA-Israel�"Iran war, advanced AI technologies named Claude AI, Maven Smart System, Gotham, Lattice, Hivemind, MQ-9 Reaper, AGM-114 Hellfire missile, Lavender, Stuxnet and Pegasus spyware were used by the US and Israel for intelligence analysis, target identification, battle simulation, AI decision support, satellite imagery, precise tracking, AI-assisted kill-chain targeting, autonomous drone and aircraft control, swarm drone operations, missile strikes, phone hacking, industrial sabotage, person identification, and real-time battlefield management. 
On the other hand, Iran employed Shahed 136 drones, Hadid-110, AI-assisted missiles, Wiper malware, Aria combat robots, and UAV swarms featuring GPS/autonomous navigation, pre-programmed targeting, radar/command-centre strikes at speeds of over 510 km/h, anti-jamming communications, algorithmic target calculation, automated data destruction and weapons operation. “Operation Epic Fury” and other attacks used FPV (First Person View) and quadrotor strike drones, thermite-based Dragon Drones, and AI-driven facial recognition, biometric surveillance, and crowd/image recognition systems, enabling the identification of individuals and civilians, including in Tehran’s official building strike and at Shajareh Tayyebeh Girls’ School in Minab, causing massive civilian deaths (168�"180 students).
AI systems like Lavender and “Where’s Daddy?” generated tens of thousands of target recommendations with an estimated 10 per cent error rate, facilitating home bombings and civilian killings. In Gaza and Libya, similar AI-assisted attacks resulted in at least 64,656 Palestinian deaths, including more than 18,000 children, with potential genocide implications, according to UN investigations and Francesca Albanese. A former Israeli colonel stated that over 2,000,000 Palestinians have been killed or injured. This system was trained using labelled datasets of Hamas members and was also capable of searching for similar behavioural patterns. It could access the phone numbers and addresses of people who were mostly civilians. Within the first six weeks after October 7, 2023, Lavender generated around 37,000 target recommendations with a 10 per cent error rate, which means thousands of civilians may have been misidentified and killed indiscriminately.
Moreover, the extreme debris crisis caused by AI-generated weapons, missiles and bombs includes asbestos, industrial chemicals, and unexploded tiny or toxic particles from these bombs or missiles. In modern warfare, these highly affect children who come into contact with unexploded ordnance. Dust from destroyed buildings can contain asbestos fibres, causing lung cancer, infections, hepatitis A cases and other respiratory diseases, while also creating environmental risks for civilians. In the Gaza war, about 39 million tonnes of debris were generated. Around 5-10 per cent of munitions fail to explode, leaving tens of thousands of unexploded bombs hidden in rubble, posing serious risks to civilian lives. Experts warn that clearing these explosives may take 14 years or more. Also, the ongoing Israel-US war against Iran is creating major debris-related risks for civilians through comparable patterns of urban warfare.
Furthermore, these systems cannot reliably determine whether a person is a combatant or a civilian because technical problems include false matches, data biases, poor lighting, algorithmic bias, cognitive bias, speed and accuracy problems, verification failures, system failures, and so on. These limitations pose serious risks to civilians and threaten their lives.
These harmful effects of AI are violating the principles of distinction, proportionality, and precaution under International Humanitarian Law (IHL), which are codified in the Geneva Conventions and their Additional Protocol I. But still now, there are no codified law under IHL specifically governing AI-based targeting mechanisms, including how they should be regulated, to what extent they should be applied, and who should be held liable if they harm civilians. There is also no established judicial case in a national or international court that has been formally filed to date.
Reports from the International Committee of the Red Cross and the United Nations caution that these unlawful automated procedures risk inaccurate targeting, accountability gaps, and reduced human oversight.
So, there should be strict adherence to IHL principles, and pre- and post-operational accountability and transparency must be ensured. Most importantly, a strict, binding law should be established under IHL to regulate the unethical use of AI and prevent indiscriminate attacks on civilians. 
Otherwise, how long will innocent civilians and children continue to die?