Negotiations over autonomous weapons continue to stall while AI-enabled systems are already reshaping conflicts in Ukraine, Gaza, and Iran.
What was supposed to happen this year was a legally binding treaty restricting weapons that can find and kill a target without a human pulling the trigger. UN Secretary-General António Guterres set 2026 as the deadline for that instrument back in his 2023 New Agenda for Peace. With the year now more than half over, no such treaty exists, and diplomats close to the negotiations say none is likely to arrive on schedule. The gap between that stalled diplomacy and what militaries are already doing on active battlefields is the real story.
A Deadline Set to Lapse
The primary forum for these talks, the UN Group of Governmental Experts on lethal autonomous weapons systems, has been meeting under the Convention on Certain Conventional Weapons since 2014. In December 2024, the UN General Assembly adopted a resolution on the issue by a vote of 166 in favour, three against, and fifteen abstentions, and a first dedicated UNGA meeting on autonomous weapons drew 96 countries in May 2025. Momentum on paper has been real. The binding text has not been followed. According to researchers who track the negotiations, reaching a legally binding protocol by the 2026 deadline would require a fundamental shift in the position of the UN Security Council’s permanent members, none of which currently supports a binding ban.
The disagreement is not just political but definitional. The International Committee of the Red Cross and most states involved describe an autonomous weapon as one that, once activated, can select and engage targets without further human intervention. Roughly thirty countries and 165 non-governmental organisations back a pre-emptive ban on any system that operates beyond meaningful human control.
The United States, by contrast, has rejected a ban outright, relying instead on its own 2020 ethical principles and a 2023 Defence Department directive that permits autonomous targeting under internal review rather than external prohibition. Where one side sees an emerging category of weapon that requires a new legal instrument, the other sees existing law as already sufficient, provided militaries govern themselves responsibly.
The Battlefields Have Already Decided
While the legal debate continues, three current wars have already moved past it in practice. In Ukraine, drones accounted for 96 per cent of Russian battlefield casualties in March 2026, according to tracking cited by security researchers, and Kyiv’s own deputy defence minister has said autonomous targeting has been partially implemented in some systems, even if fully autonomous weapons are not yet standard issue. The war has reorganised itself around drone warfare faster than any legal framework could plausibly follow.
In Gaza, Israel has used AI-assisted targeting systems known as Lavender and Habsora to help identify strike targets, with Lavender reportedly capable of approving a target within twenty seconds and, since October 2023, compiling a list of 37,000 potential individuals without full verification of their military role, according to a 2025 report from a policy research organisation tracking the war’s use of AI.
Iran has provided the most recent and largest-scale example. Under what the US military designated Operation Epic Fury, more than 13,000 targets had been struck as of April 9, 2026, with 1,000 hit on the opening day alone, according to figures released by US Central Command. Behind that pace sits the Maven Smart System, a Palantir-built platform that grew out of a 2017 Pentagon computer-vision programme and became a formal program of record in 2023. Its purpose is to compress targeting decisions that once took analysts days into a process measured in seconds.
Rather than accept an international prohibition, the United States has pursued layered domestic governance. A 2023 political declaration on responsible military use of AI drew 58 state endorsements. In Congress, senators introduced the Responsible Artificial Intelligence Defence Act this year, which would require the Pentagon to classify military AI systems by risk level, bar AI from nuclear launch decisions, and mandate that human judgment remain part of any lethal autonomous engagement. The bill has not yet passed. Its existence signals that Washington intends to keep autonomous weapons policy inside domestic legislation and internal directives rather than cede ground to a binding international treaty that would apply equally to rivals it does not trust to comply.
China’s Manufacturing Edge
Beijing has taken a parallel but distinct path, treating autonomy as a manufacturing and scale problem rather than primarily a legal one. At the 2024 Zhuhai Airshow, a Chinese state defence manufacturer displayed an entire brigade of AI-controlled armoured vehicles and drones. In January 2026, footage broadcast by the PLA’s National University of Defence Technology showed a single soldier operating a formation of 200 autonomous drones simultaneously. According to defence officials briefed on the comparison, the Pentagon’s internal concern is not that China has a conceptual lead in autonomy, but that it cannot currently match China’s industrial capacity to produce these systems at the volume a prolonged conflict would demand.
The practical consequence of failing to reach a binding instrument is proliferation beyond state militaries altogether. According to tracking by the Armed Conflict Location and Event Data Project, only ten non-state armed groups had drone capability in 2010, compared with 469 groups across seventeen countries in 2025, 58 of which used the technology for the first time that year. Groups including Colombia’s FARC and ELN have already adopted drone tactics modelled on Ukraine’s battlefield innovations. Every year the international legal architecture remains unsettled is a year in which the technology it was meant to govern becomes cheaper, more capable, and more widely available to actors with no stake in the diplomacy at all.
Guterres’s 2026 deadline will likely pass the way most UN disarmament deadlines do, quietly, without formal announcement that it has failed. The targeting systems it was meant to constrain will keep operating in the meantime, in wars that are not waiting for Geneva to agree on a definition.
