UN disarmament chief warns military AI is squeezing out human judgment

AI is shortening the path from identifying a military target to striking it, while leaving less room for human judgment, UN disarmament research chief Robin Geiss warned on October 20, 2026. At a briefing in New York, he linked that accelerating pace to greater risks of mistakes and unintended escalation—and outlined a forum to address military AI risks.

Geiss, director of the United Nations Institute for Disarmament Research, or UNIDIR, told journalists at UN Headquarters that AI was on track to crowd out human decision-making in military operations. His warning extended to the nuclear domain, where he said human judgment was also losing ground.

Faster targeting, harder-to-read systems

The responsibilities being handed to AI span several parts of military operations. Geiss described its use in intelligence analysis and target identification, as well as operating drones and coordinating cyber-attacks. His concern was not limited to a single weapon: he described a widening role for AI across military affairs.

That shift is dramatically increasing the pace of battlefield operations, he said. Alongside the shorter interval between identifying and striking a target, forces can engage an unprecedented number of targets each day. Geiss also warned that many evolving AI systems are opaque, making their workings difficult to understand.

Access is another part of the risk. As AI tools become easier to use, Geiss warned, non-state actors will find it easier to carry out increasingly sophisticated military operations. His warning therefore covered both how established forces use AI and who else can gain access to those capabilities.

Are we going to let machines target and kill human beings?

Robin Geiss, UNIDIR director, at the October 20 UN briefing

Three building blocks for military AI cooperation

Geiss said UNIDIR’s Centre of Excellence on AI, Peace and Security would launch a multi-stakeholder forum on Thursday, October 22. Covering AI in international peace and security and the military domain, it would be the first and only forum of its kind within the UN system, he said.

The forum’s stated agenda divides the work into three areas, addressing companies’ development choices, the language used in security discussions and the operational risks posed by armed groups:

• Responsible industry behavior. Develop practical, voluntary guidelines for companies designing and developing military AI tools. This part of the agenda focuses on choices made while the tools are being created, rather than only on their use in military operations.
• A shared AI security vocabulary. Clarify terms such as autonomy, safety and security. Geiss said different stakeholders currently use these concepts in very different ways, creating ambiguity that gets in the way of trust and cooperation.
• Misuse by non-state armed groups. Address the risk that these groups could use AI to support operational planning, execution and logistics. The scope reaches beyond identifying or striking targets to the work that supports an operation.

Shared concern meets geopolitical competition

Geiss acknowledged that geopolitical competition and tensions remain serious obstacles to regulating military AI. But he also described growing agreement about its risks across political divides, including among major global powers. He presented UNIDIR’s work as an effort to create the building blocks for multilateral cooperation.

The forum’s voluntary industry work sits alongside a stronger regulatory demand. The UN Secretary-General has called for tighter military AI regulation and a legally binding treaty on lethal autonomous weapons systems. Geiss said those appeals were increasingly finding support internationally. The announced forum agenda concerns guidelines and cooperation, not the adoption of that treaty.