Source: French to English Tester Published on: 2026-04-21
Source: The Conversation – France in French (3)– By Louis Perez, Postdoctoral Researcher in International Law, University Paris-Panthéon-Assas
The ongoing conflict in Iran demonstrates the advanced dependence of the US and Israeli militaries on military AI, particularly for targeting and strike planning. The bombing of a school in Minab on February 28, presented as a targeting error and causing the death of 168 civilians, mainly children, highlights legal risks, system flaws, and accountability issues.
The armed conflict against Iran launched last February 28 by Washington and Tel Aviv was quickly described as“first AI war”. A statement that is actually misleading in several respects. Not only has AI already been used intensively in recent conflicts, notablyby Israel in Gaza, but more broadly, AI, as a digital means of processing and analyzing data, has a long-standing history with armed conflicts, including the technical foundationsdate back to the Second World War.
Certainly, the Iranian situation is distinguished by the unprecedented level of sophistication of these means and by the unprecedented dependence of the armies on them. It also differs from the conflict in Gaza in that, this time, AI was deployed against a state adversary in the context of a high-intensity war. Finally, never have states soopenly communicated about their use of these systems. It is this communication combined with the dramatic consequences of certain strikes that raises questions about the compatibility of these practices with international law.
The facts: the use of AI in the war in Iran
Israel’s use of AI in its war against Hamas had beenrevealed by the newspaper+972. This media outlet had exposed what many specialists had suspected for several years. In the context of the conflict in Iran, however, it is the American authorities themselves whohave announcedtheir use of AI.
Indeed, the American military forces admitted to having used AI systems to identify and sort the list of targets at a lightning-fast speed. This process would have resulted inmore than 1,000 strikes, described as very precise, during the first twenty-four hours of the conflict. They reportedly used the system notablyMaven Smart System, a joint project using Palantir’s AI software for surveillance and data collection, coupled with the generative AI system Claude, developed by Anthropic.
However, on the first day of the war, one of the American strikestargeted a school in Minab, causing the death of about 170 civilian victims, mainly children. The United States hasacknowledged their responsibilityin this strike, presented as a mistake. The school was indeed located near a naval base of the Guardians of the Revolution. It used to be an integral part of the same complex before being separated from it. It was therefore outdated information that would have led to authorizing the strike.
Such a misunderstanding is not trivial. ManymediaandNGOquickly established the link between the school and the naval base. It was thus put forward that the American army had probably targeted this building based on outdated data by blindly following arecommendation from an AI systemwithout carrying out the verification that was required.
The legality of using AI
To what extent is the use of AI to carry out these strikes, and the mistake made, lawful under international law?
It should first be specified that AI is not prohibited as such by the law of armed conflict (LOAC, also known as international humanitarian law). For the time being, no legal rule specifically addresses the question of its legality. However, the issue does not evolve in a legal vacuum. The general rules of LOAC apply to the conduct of hostilities, regardless of the means and methods deployed.
One of these rules is theprinciple of distinctionaccording to which only military targets may be subject to attacks, and civilian persons and property must be preserved. Directly targeting a school, such as the one in Minab, in the absence of any military objective within it, therefore constitutes a clear violation of this principle. However, it is unlikely that the American military had the deliberate intention to destroy the school as such. As indicated, it is more likely a target identification error, possibly linked to an AI system trained on outdated data, dating back to the time when the building was still attached to the naval base.
Consequently, the violation is rather related to the precautionary principle. The latter notably prescribes that the parties to the conflict must do everything that is practically possible toverify that the targets to be attacked are indeed military targets. In this case, the American army does not appear to have carried out the necessary checks to ensure that the target was a school. A basic verification, like the one conducted by some media outlets, could have quickly dispelled any doubt.
It should be recalled that, during the war in Gaza,he had been reportedthat Israeli soldiers sometimes had only twenty seconds to validate a target, which raises questions about the practical possibility of effectively adhering to this principle. Concerns related to military AI often focus on the issue of autonomy and the risk that a system designates and engages a target on its own; this is the challenge posed by lethal autonomous weapon systems. However, this example shows that formally maintaining human control may be merely fictitious if the operator lacks both the time and the critical thinking needed to evaluate an algorithmic recommendation.
On the Iranian side, it should be noted that the precautionary principle was not respected either. This principle imposes obligations not only on the attacker but also requires the attacked party to take certain passive precautions: in particular, the parties mustremove civilians and civilian property from military objectives. In this case, converting a building of a naval base into a school, while keeping it in immediate proximity to the rest of the military complex, deliberately exposed this civilian facility to the risks associated with the conduct of hostilities.
What legal, political, and moral responsibilities?
Individual responsibility. The attack does not constitute a war crime.
If the attack constitutes a violation of the UCMJ, it is likely that no American military personnel will be convicted for such acts. Beyond issues of jurisdiction, the main obstacle lies in the fact that neither theviolation of the precautionary principlenor theerrorsleading to violations of IHL do not constitute war crimes under international criminal law.
The material act is well characterized, but the intentional element, that is to say the will to commit the offense, is lacking. The current regime of international criminal liability does not recognize liability for negligence in this context. However, this pragmatic approach could evolve. On the one hand, if algorithmic targeting errors multiply, the “reasonable” nature of the error will become increasingly difficult to invoke, and the conscious use of a system known for its failures could imply a form of indirect intent to target civilians. On the other hand, the law could in the future develop to punish military personnel who, through their negligence, cause the death of civilians.
The responsibility of AI companies. A tug of war between economic and political powers.
Another point of concern relates to the role of private companies specializing in AI, which today hold the majority of the technological expertise deployed on the battlefield. These companies could be held responsible when they develop faulty systems, but beyond this responsibility, a fundamental moral and political question arises regarding the sale of AI technologies for military purposes.
Just before the United States entered the war, Anthropic, which produces the Claude system, hadopposed to unlimited cooperationwith the Pentagon, notably on autonomous weapons, citing its ethical commitments and the technical reliability limits of its systems for the intended uses. The Pentagon had then accused Anthropic ofbetrayal, although its systems continue to be used by the army.
Other companies in the sector, such as OpenAI, Google, Amazon, or Microsoft, on their part, seem to collaborate unreservedly with the militaries, imposing themselvesde factolike real defense companies. It is interesting to note that companies, normally driven by profit, sometimes have more scruples in this matter than certain States, which are nonetheless guarantors of the general interest.
State responsibility. Being accountable for one’s actions and preventing future violations.
States that develop and use military AI bear a particular responsibility. In this case, the United States incurs its international responsibility for the commission of an internationally wrongful act. This responsibility will certainly be difficult to enforce in practice. But beyond that, both legal and political responsibility emerges. Under Article 1 common to the Geneva Conventions, States indeed have the obligation to respect and ensure respect for IHL. Yet, the development of military AI tends to undermine this respect, even to favor and conceal violations of the law.
Various mechanisms could curb this phenomenon, such as training military personnel on the specificities of AI systems, developing rules of engagement specific to AI, technical guarantees of reliability and transparency of systems, as well as regular testing and evaluations. Several international initiatives call for integrating such measures into new legal instruments. However, political will is lacking, particularly among the states at the forefront of the development and use of military AI.
Thus, Pete Hegseth, United States Secretary of Defense, actually seems to be acting in the opposite direction. He recently dismissed military legal advisors whom he considered asobstacles to the proper conduct of hostilitiesand hasqualifies the rules of engagement as stupid. More broadly, the United States opposes any international legal regulation of military AI. AI thus appears both as one of the drivers and the indicator of a profound erosion of IHL.
Jacques Lacan said: “The real is when you bump into it.” The Minab accident is a dramatic event that confirms the risks military AI experts have been warning about for several years and should have prompted much more reaction.
In reality, this information seems to have been overshadowed by other considerations perceived as more urgent and more visible in the context of this war, starting with thenuclear risk. The Minab accident was not the shock expected to prompt states to agree on a specific legal framework applicable to military AI. It remains to be seen whether such a shock is still possible or even desirable.
Louis Perez does not work for, advise, own shares in, or receive funds from any organization that could benefit from this article, and has declared no affiliation other than his research institution.
–ref. The use of artificial intelligence in the war in Iran: what does international law say? –https://theconversation.com/the-use-of-artificial-intelligence-in-the-war-in-iran-what-does-international-law-say-280562