When AI Pulls the Trigger: The Minab School Strike and the Limits of International Criminal Law
Amirmasoud Pourmousa
Introduction
On 28 February 2026, the United States and Israel carried out a joint military operation against Iran. Known as “Operation Epic Fury,” the campaign appears to mark a significant, and worrying, step in the growing use of artificial intelligence (AI) in warfare. According to official statements, the operation relied in part on AI-assisted systems designed to speed up target identification and selection while reducing direct human involvement. Among these systems were the Maven AI, supported by Palantir’s data analysis tools, and Claude, another generative AI model developed by Anthropic (see). These technologies were reportedly used to compile and rank potential targets within a very short time. Although the exact role of AI remains unclear, the human cost of this approach seems severe, with one strike destroying the Shajareh Tayyebeh Girls’ Elementary School in Minab, Hormozgan Province, Iran. The site was hit by three Tomahawk cruise missiles, killing 156 civilians. Among the dead were 120 children (73 boys and 47 girls) as well as 26 female teachers (see). In terms of civilian loss, this was the most serious incident in the conflict so far, and was strongly condemned by the UN experts (see).
Figure 1.1. Schematic plan of the Minab School Strike and its impact sites. (© Designed by the author)
Further investigations have suggested that the United States was likely responsible for the strike. A key issue is the Defence Intelligence Agency’s decision to classify the school as a military target. This appears to be based on the site’s earlier use as part of a nearby Islamic Revolutionary Guard Corps (IRGC) naval base. However, satellite images show (here) that between 2013 and 2016, the site was clearly separated from the base and converted into a civilian school. It was fenced off, repainted, and fitted with playground facilities, leaving little doubt about its new purpose. Taken together, these facts suggest the strike was based on outdated or inaccurate information and that the output of AI-assisted systems may have been accepted without meaningful human control (MHC). If so, this raises broader questions about the role of AI in military decision-making: (1) Can AI military use be reconciled with the rules of IHL? (2) Does this kind of mistake amount to a war crime? And (3) Who should be held responsible when critical decisions are shaped (at least in part) by AI rather than individuals?
Was the attack “Unlawful” under International Humanitarian Law?
Assessing whether the strike was unlawful requires three issues to be considered: first, the basic principles of IHL that apply to all targeting decisions; second, the practical duties that follow from those principles; and third, the conditions under which individuals can be held criminally responsible when things go wrong. At present, there is no specific rule in IHL that directly regulates the use of AI-assisted systems in warfare (see). In fact, they are often used in recent wars (e.g., in the Gaza Strip, Israel reportedly has relied on them to generate target recommendations, using AI-assisted systems such as the Gospel or Lavender). However, this does not mean its use falls outside the law. The existing rules governing the conduct of hostilities still apply, regardless of the technology involved. Whether decisions are taken by human operators alone or supported by automated systems, the same four principles remain central: (1) Necessity, (2) Distinction, (3) Proportionality, and (4) Precaution.
As Milanović stated in his opinion article (see), in the Minab case, the available information suggests the problem was not an “intentional attack on civilians,” but a “failure to properly identify the target.” Suppose that is correct, the key issue is less about necessity, distinction, or proportionality and more about precaution. This is an important point. Under rules widely accepted as part of customary international law, and reflected in Article 57 of Additional Protocol I, those who plan or approve an attack must do everything feasible to ensure the target is a legitimate military objective. They must also take all practicable steps in deciding how the attack is carried out and avoid operations likely to cause excessive harm to civilians.
The idea of what is “feasible” is, of course, not unlimited. The law does not expect perfect information in every situation. Still, it requires a reasonable effort to use up-to-date and reliable data. In this case, reliance on an AI that provided outdated information raises serious concerns, especially given that the site had been fixed and more recent information was publicly available. Satellite images (here) had already shown that the location had been converted into a civilian school, making it difficult to justify failing to recognise this change (See Figure 1.1). More broadly, the duty of precaution is not confined to the moment a strike is approved. Article 57(1) makes clear that parties to a conflict must take constant care to protect civilians and civilian objects. This is an ongoing obligation that should guide both the planning and the execution of military operations. Therefore, it appears that the attack violated IHL.
Did the attack constitute a war crime under International Criminal Law?
There is no doubt that an attack on a school is a heinous act and, in fact, a crime committed in a war. However, in legal terms, the threshold for a war crime is more complicated. It is not enough to show that IHL was breached; the act must also be committed “wilfully” and constitute a serious violation. In practice, especially under Article 30 of the Rome Statute, this requires “proof of intent and knowledge”, not a mistake or even gross negligence. This is where the argument becomes difficult.
I believe there is no doubt that the strike breached IHL. The real challenge is linking that breach to individual criminal responsibility and a war crime. The key issue is mens rea. Were those involved aware that, by attacking the military base, they might also hit a civilian structure? Or did they lack MHC in their data and genuinely were unaware that the building was a functioning school? This is the distinction between recklessness and negligence. Current international criminal law does not treat negligence (however serious) as sufficient for a war crime. If those involved did not intend to target a civilian object, the legal threshold is unlikely to be met.
Some believe, “No sane American commander would have launched an attack on a school and justified it by saying that its destruction was somehow incidental to the military advantage obtained from destroying the IRGC base. Nothing could be gained from such an act, and the US would only suffer tremendous reputational damage.” (Marko Milanovic). Still, there are multiple gaps that I can’t find answers to. Why didn’t they attack when the school was closed? How did they not check their Google Maps for once to see that there had been a school there for at least 10 years? Why didn’t they notice that there was no MHC in their data? If there are no answers, there is a possibility they “knew” or “willfully didn’t want to know,” and chose to proceed with the strike anyway. In that case, they were aware of a substantial risk that they were targeting a civilian object, and the strike could amount to a war crime.
However, let’s not forget that this is not a political science paper. Each of these assumptions needs to be supported by robust evidence; it’s not that simple to pre-assume any of these cases in a legal context. The fact is, whether the attack was the result of negligence or recklessness, it is clear that the US and Israel have long targeted Iran’s civilian infrastructure during the so-called “۱۲-day” and “Ramadan” wars. This includes attacks on Evin prison (see my article here), oil storage sites, gas fields, water infrastructure affecting 30 villages, transport networks (including railway lines, airports, roads, and bridges, e.g., Karaj B1 bridge), and communications infrastructure, such as telecom facilities and IRIB buildings (see).
Who should be held responsible?
As military operations increasingly depend on AI-assisted systems, responsibility becomes more diffuse across multiple levels of decision-making. Commanders and operators are expected to understand the limitations of the systems they use. If they continue to rely on tools known to produce outdated or unreliable information, it becomes increasingly difficult to argue that any resulting mistake was genuinely “reasonable.” In the long run, this could support arguments for forms of indirect intent or responsibility based on awareness of risk, even if such doctrines are not yet clearly established in international criminal law.
On the other hand, the use of AI-assisted systems may seem reasonable in warfare, since their algorithms are often based on the formula (identify combatants + identify civilians = combatants may be targeted; civilians must not be targeted). However, this “quantification logic” leads to excessive formalisation of human lives, and the corresponding instructions given to AI-assisted systems for operations, such as (if <x happens> then <do A> else <do B>), cannot adequately replicate human decision-making. For example, a human understands that a certain individual’s connection to a group is not strong enough to make them a legitimate target, or a person is hors de combat. Still, AI-assisted systems lack such judgment (at least for now). As a result, it fails to comply with IHL and can lead to severe consequences.
However, Concepts such as “should have known,” which appear in some areas of IHL, do not easily fit within the Rome Statute’s stricter structure. For instance, Article 8(2)(b)(ii) focuses on the intentional direction of attacks against civilian objects. The central requirement remains intention. If a strike results from outdated intelligence and no one in the chain of command realises that the target is a functioning school, attributing individual criminal responsibility becomes legally difficult. In that sense, if the incident is ultimately classified as a targeting error rather than a deliberate attack, under current legal standards, it would be very hard to identify a specific individual who meets the threshold for a war crime.
This difficulty, however, pertains only to individual criminal responsibility under the Rome Statute. It does not erase the fact that the United States remains internationally responsible for the actions of its forces. Any violation of IHL attributable to the US military (whether a targeting mistake or otherwise) engages the United States’ state responsibility under customary international law. It includes the obligation to make full reparation for the loss, like the ex gratia payment made following the downing of Iran Air Flight 655 by the USS Vincennes (See).
.
Conclusion
The Minab strike is likely unlawful under IHL for failing to apply the duty of precaution properly. Even if AI-assisted systems were involved, the core rules still apply.
A war crime requires (knowledge + intent) under Article 30 of the Rome Statute. If the incident resulted from a targeting error and those involved neither knew nor intended to hit a civilian object, “negligence” alone is insufficient. Criminal responsibility under the ICC arises only if recklessness or awareness of a serious risk can be proven.
Responsibility is difficult to pin down under the current International Criminal Law. Modern operations involve multiple decision-makers and reliance on AI-assisted systems. While responsibility may be spread across the chain of command, international criminal law still requires proof of individual intent. If this were a genuine targeting error, it is hard to attribute criminal liability to any specific person, even though the operational failure may involve several actors. That said, the United States remains responsible as a state for IHL violations committed by its forces. Regardless of individual criminal intent, this includes the obligation to pay full reparations for the loss.
Disclaimer
All blog posts, notes, commentaries, and translations published on this platform reflect the personal views and interpretations of their respective authors. They do not necessarily represent the views, policies, or positions of Law Academia or its affiliates. While we strive to maintain academic integrity and intellectual diversity, Law Academia assumes no responsibility for the accuracy, completeness, or legal validity of the content provided by individual contributors.
