AI Warfare Is No Longer a Geneva Hypothetical

When Branka Marijan attended five days of meetings in Geneva in November 2017, the debate over lethal autonomous systems could still sound like an exercise in future ethics. The forum was the Convention on Certain Conventional Weapons, the UN framework for weapons considered to cause unacceptable or indiscriminate harm. The subject was systems that could select and attack targets without a direct human decision. Much of the conversation still circled a hypothetical question: what happens if wars are one day fought by killer robots?

According to The Verge, that framing now feels dated. Not because the world has received a neatly labeled new class of machines, but because AI has been moving into parts of the military chain where the boundaries are less clean: surveillance, data processing, target selection, commander recommendations, and faster decision cycles. That is politically harder than the cinematic image of a robot holding a weapon, because the danger does not appear as one device that can simply be banned. It appears as a layer of automation inside existing systems.

That is why the phrase “autonomous weapon” is too narrow when it is reduced to a physical platform. The real question is control. Who understands what the system is recommending? Who can stop an attack? Who is accountable if an algorithmic assessment wrongly connects a pattern, object, or person with a legitimate target? The UN page for the Group of Governmental Experts on lethal autonomous weapons systems shows how long this debate has been translated into diplomatic language, and how difficult it remains to turn it into binding lines.

Technically, the central risk is not only “AI deciding by itself.” The risk is operational pressure that turns human judgment into a formality. When a system filters data, ranks threats, and accelerates response, the human in the loop can become the human at the end of the flow, approving something there is no longer time to properly examine. Such an architecture does not have to look like full autonomy to have the same practical effect: it narrows the space for doubt, verification, and accountability.

That is why red lines matter more than abstract definitions. One state can argue that a system is not autonomous because a person formally authorizes action. Another can argue that speed is necessary because an adversary is automating its own chain. A third can push development through military and industrial programs without a clear public debate. In that environment, rules based only on goodwill will not be enough. What matters are verifiable requirements: meaningful human control, a decision record, limits on targeting people, and bans on systems that the command chain cannot explain or supervise.

Here, robotics and AI are not isolated technologies. They are questions of war governance. Geneva matters as a venue, but it does not solve the problem by existing. If policy continues to move more slowly than deployment, future meetings will not be debating what might happen. They will be asking who already allowed the line to be crossed. Documents and processes, including the UN’s wider work on disarmament, matter only if they move from hypothetical formulas into operational rules that military systems must obey before use.