More than 120 states ended a week-long UN conference in Geneva on 22 November 2026 without agreeing to negotiate binding rules for autonomous weapons.
The International Committee of the Red Cross warned that governments may soon lose their ability to control the technology, as artificial intelligence reaches battlefields faster than diplomacy can respond.
A procedural agreement, not a treaty
According to The Guardian's report from Geneva, members of the UN Convention on Certain Conventional Weapons agreed to discuss an instrument regulating autonomous weapons over the next two years. Campaigners said that commitment fell well short of the urgent action required.
Dr Cordula Droege, the ICRC's director of international law, movement and partnerships, said artificial intelligence was already appearing on battlefields and possibly within weapon systems themselves.
“If we leave this unregulated, it will be much more difficult to put the genie back in the bottle,” she said.
Her warning exposes the central weakness of the Geneva outcome. A decision to keep talking preserves the diplomatic process, but it does not establish a single prohibition, create legal obligations or compel states to change weapons already under development.
A senior diplomat with knowledge of the discussions framed the meeting as a test of political intent after 13 years of debate.
“This is the only thing that is important this week. It's not really the substance of laws. What's on the agenda here, after 13 years, is the political will finally present to start to get serious about regulation of autonomous weapons systems,” the diplomat said.
What the Red Cross and UN are demanding
ICRC president Mirjana Spoljaric Egger and UN secretary general António Guterres have both called for a legally binding instrument containing clear prohibitions and restrictions. In its own appeal, the International Committee of the Red Cross said states must launch negotiations on a new protocol before it is too late to preserve meaningful human control over force.
The organisation wants autonomous weapons banned from being used against people. It also argues that such systems must behave predictably and face limits that confine their use to military objectives.
Guterres has placed the argument in explicitly moral terms. Ahead of the Geneva meeting, NPR reported his warning that allowing fully autonomous weapons to kill would cross a moral red line.
That position does not amount to opposition to every military application of artificial intelligence. It distinguishes between technology assisting human decisions and machinery acquiring the practical power to select and attack human targets without meaningful supervision.
How autonomous weapons are already being used
Autonomous weapons are military systems that act without human supervision, often through artificial intelligence or robotics. Their functions vary considerably, which has allowed governments and manufacturers to present regulation as technically premature even as operational uses multiply.
Israel's military has used AI to generate target banks in Gaza, accelerating the production of potential targets beyond what humans could compile unaided. Rights groups said the resulting methods contributed to the erroneous killing of civilians and people with tenuous connections to militant organisations.
In the Ukraine-Russia war, semi-autonomous drones have demonstrated “last-mile capability”, with AI taking control during the final approach to a target. This may appear narrower than full autonomy, but it still transfers a decisive stage of an attack to software.
The civilian cost of weak control
The dangers are no longer hypothetical. In July, a Russian drone guided by an AI system killed three civilians in Zaporizhzhia, in south-eastern Ukraine. The drone apparently intended to strike a petrol station but hit a wall instead, killing the three people with shrapnel.
In Iran, a US strike on an elementary school killed 156 civilians, including 120 schoolchildren. The attack reportedly resulted from faulty intelligence gathered by AI systems.
These cases illustrate a problem that legal debate can obscure. Human involvement elsewhere in a military chain does not necessarily prevent harm when flawed machine-generated information determines whom or what forces attack.
Swarms and proliferation raise the stakes
Anna Hehir, head of military AI governance at the Future of Life Institute, warned that further delay could produce more episodes of mass civilian harm like those seen over the previous 18 months to two years. Some future mass-casualty events, she said, could dwarf those already recorded.
Manufacturers and militaries are developing “swarm” capabilities in which hundreds or thousands of autonomous drones operate simultaneously. Hehir said an algorithmic error affecting one system could direct an entire swarm down the same erroneous path, particularly if its systems were programmed to target humans.
Proliferation creates a second challenge. Yemen's Houthi movement has used Claude to “vibe code” ballistic missiles, while armed Islamist groups in Africa's Sahel region have reportedly employed AI tools to improve drone ballistics and assist operational planning.
Waiting for major powers to settle every technical detail therefore carries its own risk. Non-state groups are already adapting accessible AI tools, while private companies and armed forces can establish operational norms through deployment rather than law.
Where should the legal red lines sit?
Droege rejected the suggestion that regulation requires an indiscriminate prohibition on every autonomous system.
“It's not about banning all autonomous weapon systems, but to set limits in an international treaty of red lines that should never be crossed,” she said. “We're very close to crossing those red lines, particularly targeting human beings through machines should be a moral red line.”
This distinction offers governments a practical basis for agreement. Predictability, restrictions on permissible targets and a prohibition on attacking people autonomously are concrete subjects for legal drafting, not abstract objections to technological progress.
For states that profess commitment to individual dignity and accountable government, the principle at stake is exacting but clear: lethal decisions require identifiable human responsibility. A machine cannot answer to a court, explain its reasoning to a bereaved family or bear legal liability for an unlawful killing.
The Geneva meeting kept negotiations alive. It did not yet demonstrate that governments are prepared to subordinate military convenience to that principle, and the next two years of discussions will show whether the promised process can produce binding law.
Autonomous weapons regulation: key questions
Did the Geneva conference agree to ban autonomous weapons?
No. More than 120 states agreed to discuss a regulatory instrument over the next two years, but they did not agree to negotiate or adopt a legally binding treaty.
Does the ICRC want every autonomous weapon banned?
No. The ICRC seeks clear international limits, including a ban on using autonomous weapons against people, predictable system behaviour and restrictions confining attacks to military objectives.
Why are swarms particularly dangerous?
Hundreds or thousands of drones could operate at once. The Future of Life Institute warns that one algorithmic error could then affect the whole swarm, multiplying civilian harm.
