For years, debates about autonomous weapons and artificial intelligence (AI)-assisted targeting remained largely theoretical. Critics warned of future harms, while some powerful countries insisted existing legal frameworks and human oversight would ensure responsible use. The absence of identifiable victims allowed many to treat these concerns as speculative.
That changed this year. Following a US strike on a girls’ school in southern Iran that killed at least 157 people, most of them students, public attention quickly turned to whether AI-assisted targeting had played a role. There was a strong indication that AI targeting was being used in the war, with some 1,000 targets struck in the first 24 hours. Although subsequent reporting suggests outdated intelligence — rather than AI itself — was likely responsible, the debate exposed deeper concerns about the quality of intelligence feeding military decisions, the pressure to generate targets rapidly and the growing reliance on AI-enabled systems operating on imperfect data. These questions became even more urgent following a joint investigation by The Independent and Airwars, which identified Abdul-Rahman al-Rawi, a 20-year-old student killed in a US strike in Iraq in 2024, as the first publicly known victim of an airstrike in which AI-assisted targeting was officially acknowledged.
For most people far from conflict zones, autonomous weapons and AI-enabled targeting may still seem abstract. However, the issue matters because the rules governing war shape how violence is used — and misused — against real people. Those rules exist precisely to constrain violence and protect civilians, and they are under pressure.
The question governments should now be asking is how AI is changing human judgment in war. As these systems become embedded across military planning and targeting, understanding how they shape what commanders see, how quickly they act, and how they exercise legal and ethical judgment are just as important as understanding the technology itself.
Military organizations are expected to seek decision advantage. AI systems promise to process vast quantities of information, identify patterns that humans might miss, generate target recommendations and accelerate what militaries call the “targeting cycle.” In increasingly complex operational environments, these capabilities appear to offer clear military benefits. Proponents often argue that AI tools may improve situational awareness, reduce cognitive burden and enable faster responses to evolving threats. But military advantage, or the promises of the technology, should not be confused with better decision making or the practical realities.
Judgment in war has never been simply about making decisions quickly. It requires weighing uncertainty, interpreting context, questioning assumptions, and balancing military necessity against legal and humanitarian obligations. AI can support those processes, but it also changes the conditions under which they occur. When AI systems produce recommendations at a pace that humans struggle to interrogate, or when operational success becomes associated with the speed and scale of targeting, the opportunity for meaningful deliberation begins to shrink. As a result, human judgment risks becoming increasingly procedural rather than genuinely reflective. This matters because the consequences extend far beyond individual strikes.
When targeting decisions are made faster, mistakes can also spread faster. Civilians are often the first to bear those consequences, particularly in densely populated urban environments where distinguishing combatants from civilians is already difficult.
For more than a decade, governments have debated these challenges through discussions on lethal autonomous weapons systems at the United Nations. A broad consensus has emerged that existing international humanitarian law applies to AI-enabled weapons and that humans must retain responsibility for decisions to use force. Yet recent conflicts suggest that affirming these principles is becoming easier than implementing them. “Meaningful human control” has become one of the defining concepts in these debates, more recently expanded to include context-appropriate human judgment and control. But invoking these concepts is not enough. The more difficult question is whether meaningful human judgment is actually possible under the operational conditions that AI is helping to create. Do commanders understand how AI-generated recommendations are produced? Can operators realistically question the outputs of increasingly complex systems while working under intense time pressure? Does the technology gradually narrow the range of options considered acceptable, or nudge certain decisions over others through system design?
These questions deserve far greater attention because AI does more than generate recommendations. As scholars and analysts highlight, it increasingly shapes the decision environment itself, what information is prioritized, which options appear most credible and how much time decision makers have to exercise independent judgment. Even when a human formally authorizes a strike, the quality of that judgment may already have been influenced by the design of the system, the data it relies upon and the pace at which decisions are expected to be made.
This is why governance must extend beyond the moment a weapon is employed. Accountability begins long before a commander authorizes the use of force. It begins with the data selected to train AI systems, the assumptions built into algorithms, the testing and validation processes used before deployment, and the institutional incentives that reward speed and efficiency. It continues through operational use and must include rigorous post-strike review and mechanisms for learning when systems fail.
States therefore have opportunities to strengthen governance now, even if negotiating new international agreements remains politically difficult. Key is to move beyond treating human judgment as a procedural requirement and instead ensure it remains a meaningful operational capability. That means providing commanders and operators with sufficient training, time, authority and institutional support to critically evaluate AI-generated recommendations rather than simply approve them.
Additionally, governments should significantly improve transparency around AI-enabled military systems. While sensitive operational information will always require protection, states should be clearer about how these systems are tested, validated, monitored and evaluated against their legal obligations. Without greater transparency, meaningful accountability becomes exceedingly difficult.
Finally, states should strengthen oversight and accountability mechanisms across the entire life cycle of AI-enabled military systems. This should include documentation of decisions made throughout the life cycle by the operators, commanders and legal advisers involved, in a form that enables genuine traceability of accountability and supports independent review, not just internal record keeping. Such documentation would also encourage a greater understanding of the role of human judgment and how it was, or was not, enabled in practice.
Abdul-Rahman al-Rawi was not a theoretical example. He was a student whose death has given a human face to debates that governments have conducted largely in the abstract for more than a decade. In the coming months, states will decide on the future of the discussions on autonomous weapons as well as ways to move forward on broader military AI dialogue. The question before them is how to ensure that human decision makers retain the capacity to exercise careful, informed and independent judgment in environments increasingly shaped by AI. Protecting civilians in the age of AI will ultimately depend on the answer.
