Artificial intelligence has entered the battlefield faster than the diplomats have entered the conference room. Autonomous targeting assistance, drone swarms and AI-driven logistics are operational realities in current conflicts, and the United Nations is now launching an international forum dedicated to military AI — an attempt to build governance around a technology that has so far been governed mainly by momentum.
The questions awaiting the forum are the hardest in modern arms control. Where must a human remain in the decision to kill? What counts as ‘meaningful human control’ when systems act in milliseconds? Who is responsible when an autonomous system errs — the commander, the programmer, the state? Our world news pages have tracked the debate from Geneva’s long-running expert talks to this new, more political stage.
Why a Forum, and Why Now
The existing expert process has spent years circling definitional questions while the technology sprinted. A dedicated forum raises the level — heads of delegation rather than technical advisers — and broadens participation beyond the arms-control club, reflecting a reality: AI capability is commercial and global, developed in civilian labs from Austin to Shenzhen, which makes Cold War-style treaty machinery a poor fit.
Military establishments are not monolithic here. Many senior officers are themselves uneasy about delegating lethal judgement and about adversaries’ systems they cannot predict — stability, the old arms-control currency, is scarce in a technology nobody fully understands.
Realism About Outcomes
Nobody expects a binding treaty soon; the forum’s plausible products are norms, red lines around nuclear command systems, transparency habits and incident-prevention channels — the connective tissue that crises need before they happen. The United Nations has performed this convening function for every destabilising technology since 1945. Military AI is next in the queue, arriving, as usual, after the fact — the task is to make ‘after’ as short as possible.
The verification problem will haunt the forum as it haunted every predecessor regime. A missile can be counted; an algorithm’s capability lives in weights and training data, invisible to inspectors and obsolete in a year. That is why the most credible proposals focus on behaviour and deployment — prohibitions on autonomous control of nuclear weapons, requirements for human authorisation of lethal force, incident hotlines — rather than on counting code. Civil society’s role is also new: the companies building the frontier models are not defence contractors in the old mould, and their own safety commitments have become diplomatic facts. The forum’s success metric is modest and real: whether, five years from now, the world’s militaries share a vocabulary for restraint and a channel for the day an autonomous system does something nobody ordered. Arms control has always begun exactly there.
Cynics will note that forums are where urgency goes to be minuted. The historical counter is that every arms-control achievement began as exactly such a room — people mandated to talk about the unspeakable until it became, slowly, governable.
