Introduction
The conflict between Israel and Palestine has deep historical roots and dates back more than a century (Global Conflict Tracker). From the first conflict in 1946 to the recent hostile activities, the violence has persisted. The historical and ideological conflict between Israel and Palestine has seen innovations on both sides to inflict more damage on the other. Recently, new technologies such as Artificial Intelligence have started to play a bigger role in modern warfare. The ongoing conflict in the Gaza Strip shows that AI is being used more and more in military operations, with the Israeli military having used Artificial Intelligence Decision Support Systems (AI-DDS) in the Gaza Strip (IE Med, 2025). Systems like ‘Lavender’ and ‘Gospel’ help the Israeli Defence Forces (IDF) to pick, track, and target their victims, supporting human decision-making through surveillance, big data, and facial recognition. While AI-assisted targeting systems make military operations faster and more accurate for the IDF, there are many legal and ethical issues surrounding their use.
This article examines whether these AI Systems respect the core principles of International Humanitarian Law (IHL). The primary purpose of IHL is to restrict the means and methods of warfare and ensure the protection and humane treatment of persons who are not, or no longer, taking a direct part in hostilities (Melzer, 2019). The main goal of this article is to determine whether the use of AI-assisted systems in warfare complies with IHL and its core principles. The focus will be on the conflict between Israel and Palestine, the use of these systems in the Gaza Strip, and their compatibility with the current legal framework. It should also be noted that Israel has not ratified the main sources of IHL, Additional Protocols I and II, but the core principles of IHL still apply under customary international law (ICRC, Customary IHL).
Overview of AI Systems in Warfare
The IDF utilises AI systems for both combat and support functions. The AI systems employed by the IDF for targeting fall under the decision support systems (DSS), technologies that assist commanders in their decision-making process (IE Med, 2025). AI-DSS is designed to help humans make complex decisions by using large amounts of data to identify patterns, present relevant information, or provide recommendations.
One of the AI-assisted systems that IDF uses in Gaza is called Lavender. Reportedly, Lavender was created to help identify people in Gaza who are suspected of being operatives in the military wings of Hamas or other armed groups (Andersin, 2025). The system assigns a score from 1 to 100, indicating the likelihood that a person is a member of one of these groups (Future for Advanced Research, 2024). It uses machine learning to go through huge amounts of surveillance data that has been collected on almost everyone in Gaza, including the phone contacts, social networks, facial recognition from cameras, drone footage, and even data from military checkpoints (Andersin, 2025). Lavender’s influence on the military’s operations was so extensive that the AI’s outputs were treated as human decisions (+972 Magazine, 2024). It seems like the system is not really looking for clear evidence of fighting or planning attacks, it is more about finding patterns (Future for Advanced Research, 2024).
This kind of targeting is very problematic under IHL. This approach means that Lavender identifies potential targets based on statistical likelihood rather than real evidence. One principle of IHL states that the parties to the conflict must always distinguish between combatants and civilians (Article 48, ICRC). However, if people are selected for targeting solely because of their contacts or habits, rather than their direct participation in hostilities, there may be a risk of false identifications. One anonymous Israeli intelligence officer stated that the army granted sweeping permission for officers to adopt the target list generated by Lavender without further examination, despite the system having an error rate of about ten percent (+972 Magazine, 2024).
Interestingly, Lavender is also used alongside other systems such as The Gospel, which helps choose locations for airstrikes, and another tool, ‘Where’s Daddy?’ to track people to their homes so strikes can happen when they are there (Andersin, 2025).
The Gospel system is used to support military decisions; instead of targeting people like Lavender, it focuses on buildings and structures, known as ‘fixed targets.’ They are mostly the residential homes where Hamas members live (IE Med, 2025). It works by analysing various signs to determine which locations might be important for the military. For example, it looks at buildings with extra protection, places where many people gather, or spots that look different compared to older satellite images (Future for Advanced Research, 2024). Gospel also uses specialised imaging technology to determine which materials were used in a building, and whether it is natural ground or a manmade structure, such as tunnel entrances. It also looks for signs of movement, like heavy equipment or weapons being moved around. This could show that the place is being used for military purposes (Future for Advanced Research, 2024). On the other hand, ‘Where’s Daddy?’ is an additional automated system that was used to track the targeted individuals and carry out bombings when they had entered their family residences. Consequently, the Israeli army systematically attacked the targeted individuals while they were at home, because it was easier to locate them (+972 Magazine, 2024).
Both systems are supposed to help commanders make decisions by processing large amounts of data and offering recommendations; they are not described as fully autonomous systems (Ploughshares, 2024). However, the use of these systems in practice raises many questions. For example, Lavender had an error rate of around ten percent, meaning that one in ten people marked as targets had no proven link to armed groups (IE Med, 2025). There are also wider issues with how the systems were trained and used. The data could be flawed, the models might carry hidden biases, and the military seems to have accepted a high tolerance for mistakes (IE Med, 2025). This can lead to unfair or even discriminatory outcomes, such as targeting someone solely based on their contacts or behavioural patterns (Batallas, 2024).
So, even though Lavender and Gospel are officially not autonomous weapons, the way they were/are used in the Gaza Strip shows a different pattern. In practice, they operate with a high level of autonomy, with humans mostly approving what the algorithm has already decided. This can blur the line between decision-support systems and fully autonomous weapons.
Compatibility with the Principles of IHL
The primary purpose of IHL and its core principles is to restrict the means and methods of warfare and ensure the protection and humane treatment of persons who are not, or no longer, taking a direct part in hostilities (Melzer, 2019). The principle of distinction is a cornerstone of IHL, grounded in the recognition that ‘the only legitimate object which states should endeavour to accomplish during war is to weaken the military forces of the enemy’ (St. Petersburg Declaration, Preamble).
Additional Protocol I further states that the parties to the conflict must at all times distinguish between civilian objects and military objectives and direct their operations only against the military objectives (Article 48, AP I). According to Article 51(3) of the Additional Protocol I, civilians are protected from being attacked unless, and only for as long as, they take a direct part in hostilities. This rule is strict and requires the military to be very sure someone is actively involved in the fighting at that particular moment. It is not enough to just assume based on who someone knows or where they live, just like the AI-DDS systems do now. The responsibility is on the attacker to prove that the person is a lawful target.
IHL also clearly prohibits indiscriminate attacks. As Nils Melzer explains, these attacks are intended to strike military objectives, civilians, and civilian objects without distinction, either because they are not, or cannot be directed at a specific military objective or because their effects cannot be limited as required by IHL (Melzer, 2019). The ICRC has also clearly stated that any new technology used in war must comply with the rules of IHL. It is not only about how states choose to use it, but the system’s capacity to operate within the law (ICRC, 2019).
The way AI systems like Lavender operate violates these rules. Reportedly, the Lavender software analysed information collected on most of the 2.3 million residents of the Gaza Strip through a system of mass surveillance, then assessed and ranked the likelihood that each particular person is active in the military wing of Hamas (Future for Advanced Research, 2024). Lavender assigns people ‘suspicious scores’ based on their behaviour or who they are connected to, rather than on actual proof that they are parties to the conflict (Future for Advanced Research, 2024). According to the sources, Lavender gives almost every single person in Gaza a rating from 1 to 100, expressing how likely it is that they are militant (+972 Magazine, 2025).
According to the soldiers who worked with the system, Lavender occasionally made mistakes while identifying the targets. This misidentification included police officers, civil defence workers, relatives of militants, and residents who happened to share the same first or last name as militants, or they were using a device that previously belonged to a Hamas member (+972 Magazine, 2025). This means that the system is making guesses; it uses patterns and not facts. The scale could also be alarming. Reportedly, Lavender flagged around 37,000 people; if even 10% of those were wrongly identified, that would be thousands of possible innocent people (IE Med, 2025). These concerns become even more serious when human reviewers are given only 20 seconds to approve the targets, which shows that the process lacks real verification (+972 Magazine, 2024). Moreover, there is no requirement to check why the machine made that choice or to examine the raw intelligence data on which it is based (+972 Magazine, 2024).
As a result of this technology, thousands of Palestinians, mostly women and children, and others not involved in the fighting were wiped out by Israeli airstrikes. This clearly violated the principle of distinction under the IHL. Moreover, the army systematically attacked the lawful targets in their private homes, alongside their families. According to the +972 Magazine, several sources also stated that, in the case of systematic assassination strikes, the army made the active decision to bomb suspected militants inside civilian households from which no military activity took place, and this choice was a reflection of the way Israel’s system of mass surveillance in the Gaza Strip is designed (+972 Magazine, 2024). It becomes evident that if systems like Lavender are making decisions, and humans are just approving them without adequate oversight and consideration, then that challenges the very core of the principle of distinction. Civilians may be treated as threats simply because a machine detects a pattern, and this approach does not meet the legal requirements of IHL. Human Rights Watch also stated that the use of AI to inform military targeting decisions has risked violating IHL concerning the distinction between military targets and civilians and the need to take all feasible precautions before an attack to minimise the civilian harm (Human Rights Watch, 2025). According to Human Rights Watch, AI-DSS technology reflects the biases of its developers; they are often built using incomplete data, and technically, are incapable of showing how or on what data or basis targeting decisions were made.
Conclusion
The use of AI-assisted targeting systems like Lavender and Gospel in Gaza raises many complex questions in terms of International Law. While these systems are meant to make military operations faster and more efficient, their use in practice shows that this ‘efficiency’ might come at a very high human cost. The reports and articles about this issue show that people in Gaza have been targeted based on patterns, behaviours, or associations, not on clear proof of participation in hostilities. When thousands of people are added to a kill list by a machine, and mistakes are known but still accepted, it is hard to see how the core principles of international law are respected and protected.
As modern technology and artificial intelligence evolve, the use of AI in warfare is also growing, and this will also shape how such systems are used in the future. If the international community continues to allow this kind of targeting without proper checks and balances, the principles of the IHL risk becoming less legitimate and enforced.
Bibliography
Abraham, Yuval, ‘A mass assassination factory: Inside Israel’s calculated bombing of Gaza.’ (November 30, 2023). https://www.972mag.com/mass-assassination-factory-israel-calculated-bombing-gaza/ accessed November 24, 2025.
Andersin, Emelie, ‘The Use of the ‘Lavender’ in Gaza and the Law of Targeting: AI-Decision Support Systems and Facial Recognition Technology,’ (May 23, 2025).
Batallas, Carlos, ‘When AI meets the Laws of War,’ (October 3, 2024), https://www.ie.edu/insights/articles/when-ai-meets-the-laws-of-war/ accessed November 24, 2025.
Cassidy, Jennifer, ‘Gospel truth: Israel turns AI into civilian killing machine in Gaza’ (2024). https://www.trtworld.com/opinion/gospel-truth-israel-turns-ai-into-civilian-killing-machine-in-gaza-16250446 accessed November 24, 2025.
Center for Preventive Action, ‘Israeli-Palestine Conflict,’ (Global Conflict Tracker, May 28, 2025) https://www.cfr.org/global-conflict-tracker/conflict/israeli-palestinian-conflict accessed November 24, 2025.
Future Center for Advanced Research and Studies, ‘Blind Technology: Israel’s AI Deployment in Gaza and Lebanon Wars,’ (October 7, 2024), https://futureuae.com/en-US/Mainpage/Item/9735/blind-technology-israels-ai-deployment-in-gaza-and-lebanon-wars accessed November 24, 2025.
Human Rights Watch, ‘A hazard to Human Rights,’ (April 28, 2025) https://www.hrw.org/report/2025/04/28/hazard-human-rights/autonomous-weapons-systems-and-digital-decision-making accessed November 24, 2025.
ICRC, Customary IHL Database https://ihl-databases.icrc.org/en/customary-ihl/v1 accessed November 24, 2025.
ICRC, Protocol Additional to the Geneva Conventions of 12 August 1949 (API) https://ihl-databases.icrc.org/en/ihl-treaties/api-1977 accessed November 24, 2025.
ICRC, St. Petersburg Declaration, https://ihl-databases.icrc.org/en/ihl-treaties/st-petersburg-decl-1868/declaration accessed November 24, 2025.
ICRC, Artificial Intelligence and Machine Learning in Armed Conflict: A Human-Centred Approach. (Geneva, June 6, 2019).
Marijan, Branka, ‘AI Targeting in Gaza and Beyond,’ (March 26, 2024), https://ploughshares.ca/ai-targeting-in-gaza-and-beyond/ accessed November 24, 2025.
Melzer, Nils, International Humanitarian Law: A Comprehensive Introduction (ICRC 2019).
Noah, Sylvia, ‘The Israeli Military’s Use of AI in Gaza: Operational Efficiency at the Cost of Humanity,’ https://www.iemed.org/publication/the-israeli-militarys-use-of-ai-in-gaza-operational-efficiency-at-the-cost-of-humanity/ accessed November 24, 2025.
