The Kill-Switch Talks Are Stalling in Geneva
A UN weapons convention met this week to regulate autonomous killing machines. The U.S. and Russia may have just buried it.
By The Defense Desk
In a conference room in Geneva, the 128 states party to an international weapons convention spent a week trying to agree on a single non-binding document — not a treaty, not a ban, just a text that might one day lead to rules governing machines that can select and kill human targets on their own. By Friday, even that modest goal was slipping away. [1]
The talks were held under the Convention on Certain Conventional Weapons (CCW), the United Nations framework that regulates or bans weapons deemed excessively injurious or indiscriminate. The 128 member states were deciding whether consensus could be reached on language that might usher in formal negotiations on prohibitions and regulations for Lethal Autonomous Weapons Systems, or LAWS. [1]
They could not get there — at least not cleanly. A majority of countries expressed support for a draft text, but the United States and Russia proposed late changes and deletions to it, according to two diplomatic sources. [1]
The distance between a majority and a consensus is, in diplomatic terms, everything. Agreement without unanimity is not agreement at all — it is just a number, and numbers do not bind anyone to anything.
"The window is getting smaller to getting an agreement," one diplomatic source told Reuters. A second warned that diplomats may run out of time and that key provisions discussed over several years could be dropped, running the risk of no deal emerging. [1]
The sticking points are not abstract. Differences remain over language — specifically the use of words such as "lethality" and a section related to exercising human judgment and control, according to one of the sources. [1] These are not semantic quibbles. Whether a treaty text says "lethality" or sidesteps the word determines whether signatories must ever confront the core question: who is responsible when an algorithm decides to shoot?
The U.S. position has hardened over the past six months, a senior diplomat said, with Washington favouring something more akin to guidelines than a legally binding international treaty as it says international humanitarian law already provides a framework to govern the conduct of warfare. This position is supported by states including Russia, India and Turkey. [1]
This is the oldest argument in arms control: that existing rules are sufficient, that new law would constrain more than it would protect, that flexibility in war is a feature rather than a flaw. The counter-argument is equally durable — that law without specificity is not law at all.
At least six countries received formal diplomatic approaches from Washington to back its position on the draft text, two diplomats said. This showed U.S. engagement to try to reach consensus on the issue, one of them said. Others read it differently: as a coordinated effort to dilute the text before it could gain traction. [1]
On the other side of the table, Brazil, Ireland, and several African states have pushed for a legally binding treaty, arguing that safeguards are needed to ensure meaningful human control over the weapons systems in question. [1] The phrase "meaningful human control" has become the battle line of the entire debate — three words that sound uncontroversial until you try to define them in a legal instrument that militaries would actually be required to follow.
The urgency is not hypothetical. Concerns have been mounting about the role of semi-autonomous weapons that can identify and engage targets with varying degrees of autonomy in active conflict zones including Ukraine, Sudan, and the Middle East. [1] These are not weapons being designed in a lab for some future war. They are operating now, in real theaters, under legal frameworks that were written before the technology existed.
Human Rights Watch and the Stop Killer Robots campaign group have warned that weapons systems — some assisted by artificial intelligence — may behave unpredictably, raising the risk of fatal errors. [1] States agree that international humanitarian law applies to LAWS, but specifically binding international standards for such systems remain virtually non-existent. [1]
There is a certain architectural irony to this moment. The systems that diplomats are struggling to regulate were built to operate faster than human decision-making. The negotiations meant to govern them move at the opposite pace — years of careful language, and still no binding word.
In August, UN Secretary-General Antonio Guterres and Red Cross President Mirjana Spoljaric warned that the world was nearing a moral red line where autonomous weapons could kill without human involvement. [1] That warning arrived weeks before Geneva, and it did not appear to shift the calculus for the delegations most resistant to a binding deal.
The Russian mission in Geneva did not comment. The U.S. mission did not immediately respond to a request for comment. [1] The silence from both capitals is itself informative — neither government has found it politically useful to publicly defend its position in a forum where the stakes are framed in moral terms.
What Geneva reveals is not a failure of diplomacy so much as a failure of appetite. The will to regulate is distributed unevenly among the countries most capable of building these systems. The countries least capable of building them are, in many cases, the ones most likely to be targeted by them.
If the talks collapse without a text, the years of negotiation spent crafting language around "lethality" and "human judgment" will be shelved, and the machines will keep running. The window, as one diplomat put it, is getting smaller. It is not yet closed — but Geneva this week offered a clear view of who is holding it open and who would rather it shut. [1]