The smoke rising from the rubble of the Minab school on February 28, 2026, became an indelible mark on the history of automated warfare, serving as a visceral reminder of what happens when machine speed outpaces human oversight. During a high-tempo U.S. military operation in Iran, the strike intended for a nearby munitions depot instead leveled a primary education facility, claiming the lives of 120 children and 26 staff members. While initial investigations focused on whether the artificial intelligence system suffered a technical glitch or a software bug, such a narrow inquiry fails to grasp the broader systemic failures at play. The disaster at Minab was not merely an isolated technical error but the predictable outcome of a sociotechnical system that prioritizes operational efficiency over the deliberate, often slow-moving requirements of international humanitarian law. By viewing AI as a standalone tool, military planners ignore the complex web of human operators, technical infrastructure, and organizational workflows that define how these systems function in the real world. This catastrophe illustrates a growing rift between the rapid development of targeting technologies and the established legal frameworks designed to protect non-combatants in times of conflict. Understanding this event requires moving beyond the “black box” of the algorithm to analyze the institutional pressures and technical limitations that allowed such a massive failure of distinction to occur during an active combat engagement.
The Sociotechnical Reality: Beyond the Algorithm
Viewing artificial intelligence as a purely technical solution ignores the reality that these systems are deeply embedded within human organizational structures and military cultures. When a military integrates AI into its targeting cycle, it is not simply adding a faster calculator; it is transforming the entire process of how information is gathered, analyzed, and acted upon. This sociotechnical perspective reveals that AI does not necessarily fix human errors but often provides a more efficient pipeline for existing institutional biases and intelligence failures to reach the battlefield. In many cases, the technical “efficiency” of an AI system acts as a mask for the erosion of rigorous verification protocols that were previously standard practice. By focusing only on the performance metrics of the code, commanders risk losing sight of the human and organizational checks that are supposed to act as a buffer against tragedy. The real challenge lies in the fact that these systems are designed by engineers but operated by personnel under extreme psychological stress, creating a volatile environment where technical suggestions are frequently accepted without the necessary skepticism or situational awareness required for high-stakes decisions.
The concept of “friction” has long been a foundational element of military theory, acting as the natural resistance that forces commanders to pause, double-check, and deliberate. In the context of international humanitarian law, this friction is not a hindrance but a vital legal safeguard that provides the time and space necessary to verify a target’s legitimacy and assess potential civilian harm. Artificial intelligence is explicitly designed to eliminate this friction, streamlining the transition from intelligence gathering to kinetic action in the name of maintaining a competitive advantage. However, when this deliberation time is stripped away, the human ability to intervene and prevent a strike based on faulty data is severely compromised. The desire for a “seamless” kill chain essentially removes the “human-in-the-loop” functionality that is legally required to ensure that every attack adheres to the principles of necessity and proportionality. By the time an operator realizes an error has been made, the strike is often already in progress, leaving the legal and ethical framework of the military in a constant state of catch-up with its own technology.
Statistical Probabilities: The Mirage of Precision
Modern military targeting systems are increasingly reliant on Large Language Models and probabilistic machine learning architectures that do not function through rigid, deterministic logic. Instead of following a strictly defined “if-then” set of rules, these AI systems operate by identifying statistical patterns and predicting the most likely outcome based on historical training data. While this allows for the processing of immense datasets, it also introduces a fundamental element of uncertainty that is often obscured by the sleek user interfaces presented to military analysts. These systems are optimized for pattern completion and fluency rather than factual truth or legal nuance, meaning their outputs are always estimations rather than certainties. When an AI system presents a target with a high confidence score, it is not providing a verified fact but a statistical guess that may be influenced by flaws in the underlying data. This probabilistic nature is fundamentally at odds with the high standards of verification required by the laws of armed conflict, which demand that every feasible precaution be taken to avoid misidentifying a civilian object as a military objective.
The phenomenon known as “hallucination,” where an AI generates false or misleading information but presents it with absolute certainty, poses a grave threat to civilian safety in modern combat zones. In a civilian context, an AI hallucination might result in a harmlessly incorrect answer to a trivia question, but in a targeting environment, it can lead to the misidentification of a school or hospital as a legitimate high-value target. Because the AI does not have a conceptual understanding of the world—only a mathematical understanding of data relationships—it cannot realize when its suggestion is nonsensical or dangerous. Human operators, conditioned to trust the precision of high-tech systems, often fall victim to automation bias, assuming that the machine’s output is the result of a deeper analysis than any human could perform. This creates a dangerous feedback loop where the machine’s “confidence” in its statistical prediction is mistaken for the absolute verification of a target’s identity. The result is a system that can generate highly persuasive, yet fundamentally incorrect, targeting orders that bypass the critical thinking of the military personnel who are supposed to be the final arbiters of legal compliance.
Choice Architecture: The Erosion of Human Judgment
The way military software interfaces are designed plays a significant role in how decisions are made under pressure, a concept known as “choice architecture.” In the high-stakes environment of 2026, the user interfaces of targeting platforms are often optimized to encourage rapid-fire decision-making, presenting potential targets in a streamlined, “gamified” format that can mask the gravity of the consequences. When an AI system prioritizes certain information while burying conflicting reports, it effectively nudges the operator toward a specific conclusion before they have even begun their own analysis. This architecture exploits human psychology, making it much easier to hit “accept” on a machine-suggested target than it is to challenge the system and demand more data. The pressure to maintain the operational tempo of modern warfare means that analysts rarely have the luxury of time to dig through the layers of reasoning that led the AI to its conclusion. Consequently, the role of the human operator is reduced from a thoughtful judge to a mere “rubber stamp” for automated recommendations, a shift that fundamentally undermines the legal obligation to exercise meaningful human control over lethal force.
This systemic prioritization of speed over accuracy is further exacerbated by the sheer scale of the operations that AI allows a military to conduct. When a targeting system is identifying and processing thousands of potential objectives in a single day, the human capacity to review each one becomes mathematically impossible. Analysts are forced to rely on summaries and confidence scores rather than raw intelligence, creating a situation where the nuanced details of a target’s surroundings—such as the presence of a school next to a warehouse—are easily overlooked. This “scaling up” of warfare creates a disconnect between the tactical level of the individual strike and the strategic legal obligations of the military as a whole. The loss of human judgment is not a bug in the system but a byproduct of a design philosophy that views human deliberation as a bottleneck rather than a safeguard. As these technologies become more pervasive, the standard for what constitutes “reasonable” human oversight continues to shift, potentially lowering the bar for legal accountability and increasing the risk of widespread, systemic civilian harm that cannot be traced back to a single human decision.
Intelligence Decay: The Peril of Stale Data
The catastrophic failure at the Minab school was not just a failure of AI logic but a failure of the data infrastructure that fed the system, highlighting the dangers of using outdated or “stale” intelligence. In the years leading up to the incident, analysts had repeatedly warned that the facility records for the region were increasingly inaccurate, with many buildings having changed their primary function without being updated in the central databases. However, the hunger for actionable targets in a high-intensity conflict created a culture where the volume of data was prioritized over its freshness or accuracy. When this unreliable, stale information was fed into the AI targeting engine, the machine did not have the capability to question the age of the data or verify if a “warehouse” was still a warehouse or had since been converted into a school. The AI essentially supercharged the errors within the database, turning outdated records into immediate, high-priority strike orders at a speed that bypassed the manual verification processes that might have caught the discrepancy.
This reliance on aging data highlights a structural vulnerability in AI-enabled warfare: the machine can only be as accurate as the information it is given, yet it operates at a speed that makes thorough data validation nearly impossible. In a dynamic combat environment, the “ground truth” can change in a matter of hours, but the databases used to train and inform AI systems often lag by months or even years. When military leadership continues to use these systems despite knowing that the underlying intelligence is flawed, they are essentially automating negligence. The machine does not possess the common sense to look at a satellite feed and notice the presence of school children if its database tells it the building is a military storage site. This creates a “data-reality gap” that AI is currently unable to bridge, leading to a situation where strikes are launched based on a digital phantom rather than the actual situation on the ground. The tragedy at Minab proved that high-tech processing power is no substitute for accurate, timely, and human-verified intelligence, especially when the stakes involve the lives of hundreds of non-combatants.
Legal Strain: Precautions and the Principle of Distinction
The integration of high-speed AI targeting systems places an immense strain on the Principle of Precautions, a cornerstone of international humanitarian law that requires combatants to do everything feasible to verify that targets are not civilian objects. Relying on an AI system that is known to use statistical probabilities and potentially outdated data represents a significant departure from this legal duty. If a military force is aware that its AI frequently “hallucinates” or that its intelligence databases are stale, it cannot reasonably claim to have taken all feasible precautions when a strike based on that system leads to civilian casualties. The legal standard of “feasibility” must evolve alongside the technology; if a tool increases the speed of targeting, it should also be used to increase the rigor of verification. Instead, the current trend shows AI being used to bypass the very precautions it was meant to enhance, creating a legal environment where the rush to strike outweighs the duty to protect. The failure to reconcile the speed of automated systems with the slow, meticulous requirements of legal verification is leading to a crisis of compliance on the modern battlefield.
Similarly, the Principle of Distinction, which mandates that parties to a conflict must at all times distinguish between civilians and combatants, is being eroded by the indiscriminate nature of high-speed, AI-driven targeting. When strikes are launched based on algorithmic patterns rather than individual target identification, the risk of “pattern-matching” civilians into the category of combatants increases dramatically. This reckless disregard for the nuances of human behavior can lead to strikes on groups of people who appear to be acting in a suspicious manner according to a machine’s training data, even if they are simply civilians trying to survive in a war zone. Using technology to act on unverified or probabilistic data moves the act of killing from a deliberate, legally justified decision to a statistical gamble. This shift threatens to normalize a level of “collateral damage” that was previously unacceptable, as the convenience of the technology begins to dictate the limits of legal constraint. Without a clear reassertion of the Principle of Distinction in the age of AI, the protections afforded to civilians under international law will continue to be sacrificed in favor of operational efficiency.
Accountability and the Reasonable Commander Standard
One of the most concerning aspects of AI-enabled targeting is the “diffusion of responsibility” it creates, making it nearly impossible to hold individuals accountable when international law is violated. Because the path from data collection to a strike order involves dozens of human actors and millions of lines of code, pinpointing exactly where a failure occurred becomes a monumental task for legal investigators. This “accountability gap” allows military organizations to blame technical glitches or “system errors” rather than identifying the human decisions—such as the decision to use stale data or to ignore analyst warnings—that led to the tragedy. Without a clear chain of responsibility, the deterrent effect of international law is weakened, as commanders may feel shielded by the complexity of the systems they use. The lack of transparency in how AI models reach their conclusions, often referred to as the “black box” problem, further complicates the ability of international courts to determine if a strike was a legitimate mistake or a war crime.
To address these challenges, the international community began advocating for a redefined “reasonable commander” standard that specifically accounted for the use of automated systems. It was determined that a commander could no longer claim ignorance of a system’s technical flaws as a defense; instead, the legal burden shifted toward ensuring that any AI used in combat was subject to rigorous, ongoing testing and human oversight. International legal bodies suggested that the use of high-speed targeting should be prohibited in areas with high civilian density unless real-time, human-verified intelligence could be provided. Actionable steps were proposed to mandate “explainability” in military AI, requiring that every machine-generated target come with a transparent audit trail of the data points and logic used to justify the strike. These reforms aimed to close the accountability gap by ensuring that humans remained legally and morally responsible for the outcomes of automated warfare. By prioritizing the preservation of human judgment over the allure of machine speed, these initiatives sought to restore the integrity of international humanitarian law in an era of unprecedented technological change.
