A UN conference in Geneva, scheduled for 16 to 20 November and attended by 128 countries, has been identified as a critical juncture in the global effort to regulate autonomous weapons. The core question on the table is whether nations will agree to begin formal, legally binding negotiations on AI-guided weapons systems, or whether that process will be delayed once again by a small group of powerful states.
The Governance Gap That Keeps Growing
The definition matters here. An autonomous weapon, as agreed by the expert working group preparing for the Geneva conference, is any system “capable of identifying, selecting and engaging a target without the need to refer the issue to a human being.” That is not a hypothetical category. These systems are already in use.
In July, Ukraine reported that Russia had deployed an AI-guided drone that struck a propane tank at a fuel station in Zaporizhzhia. Ukrainian President Volodymyr Zelenskyy has publicly warned about the pace of AI weapons development, calling for the war to end before AI systems come to dominate the battlefield. The technology is not waiting for the diplomacy to catch up.
Discussions inside the UN have been under way since 2014, within the Convention on Certain Conventional Weapons. What makes the November conference different is that states have already agreed by consensus to elements of a draft treaty text. The groundwork, in other words, exists. The Ireland-led coalition of 50 states, which includes France, Germany, Italy, Brazil and Mexico, has proposed a concrete mechanism: a time-limited expert group tasked with negotiating rules across three sessions in Geneva in 2027, totalling 25 days, with a draft protocol due by the end of that year. The ICRC has reported that more than 70 states support beginning negotiations on a legal instrument. A separate UN disarmament committee vote saw 163 states signal readiness to launch negotiations.
Those numbers are not small. The opposition, however, is strategically positioned.
Who Is Blocking, and How
Russia, the United States and Israel are among the states pushing against any commitment to a legally binding timetable. The US position, broadly, is that existing international humanitarian law provides a sufficient framework, and that individual nations should be left to impose their own controls. The Pentagon has established its own autonomous weapons working group, but that is an internal process, not an international one.
Anna Hehir, head of military AI governance at the Future of Life Institute, has described the likely blocking strategy plainly: states will use language that appears to signal progress while actually advocating for more years of non-binding discussions. Her phrase for it is precise: “smokescreens to keep autonomous weapons systems unregulated and states unaccountable for their use.”
The UK has not backed the Ireland-led statement and has previously argued that existing humanitarian law is adequate. France, by contrast, has welcomed the UN experts group proposals as a sufficient basis for a negotiating mandate that would ban completely autonomous weapons and regulate systems with autonomous functions.
The gap between those two positions is not technical. It is political.
A poll commissioned by the Future of Life Institute found overwhelming public support for a global treaty on AI weapons in France, Germany and the UK, alongside scepticism that any such treaty would actually be respected. That combination, broad public support paired with low institutional trust, is a familiar feature of arms control debates.
What This Moment Reveals About AI Governance
Here is what most coverage of this story misses. The Geneva conference is not primarily about autonomous weapons. It is a test case for whether international governance of AI is possible at all.
The pattern visible in Geneva is the same one that appears across AI regulation more broadly: technology develops faster than institutions can respond, powerful actors resist binding constraints, and the gap between what is technically possible and what is legally prohibited widens. The difference with autonomous weapons is that the stakes are measured in human lives, and the errors are irreversible.
The Minab case illustrates this concretely. On 28 February, a US bombing of a school in Minab, Iran, killed at least 175 people. The targeting process included input from the Maven AI system operated by Palantir. Maven did not flag that the site had been a school for ten years. Neither did the personnel assigned to review the AI-generated target data. Speed was gained. Accuracy was not.
Human Rights Watch and Stop Killer Robots have argued that AI-assisted weapons systems have demonstrated unpredictable behaviour, raising the risk of fatal errors and battlefields that slip beyond human control. The ICRC and the outgoing UN Secretary General António Guterres warned jointly that the world is “nearing a moral red line where autonomous weapons could kill without human involvement.” Pope Leo XIV, in a direct intervention timed to influence the conference, stated that decisions affecting human life and the use of force must always remain subject to human oversight, and called on states meeting in Geneva to take concrete steps toward banning lethal autonomous weapons.
The question Geneva will answer is not whether autonomous weapons are dangerous. That is already acknowledged. The question is whether the international system can move from acknowledgment to obligation.
In Short
A UN conference in Geneva is the closest the world has come to launching formal negotiations on legally binding rules for autonomous weapons. More than 163 states have signalled readiness to negotiate. A coalition of 50 nations has put a concrete proposal on the table. The ICRC, the outgoing UN Secretary General and a broad coalition of civil society organisations are calling for action now. Russia and the US are pushing to keep the process non-binding and open-ended. What happens in Geneva will not just determine the future of AI weapons. It will reveal how much the international community is actually willing to constrain the technology it has already deployed.
Based on reporting from The Guardian.