AI Policy Corner: AI Policy Corner: The UN and the Future of Military AI Governance

✍️By Ahmet Ergurum

Ahmet is a PhD candidate in Political Science at the University of Wisconsin-Milwaukee, as well as a Graduate Affiliate at the Governance and Responsible AI Lab (GRAIL) at Purdue University.

📌 Editor’s Note: This article is part of our AI Policy Corner series, a collaboration between the Montreal AI Ethics Institute (MAIEI) and the Governance and Responsible AI Lab (GRAIL) at Purdue University. The series provides concise insights into critical AI policy developments from the local to international levels, helping our readers stay informed about the evolving landscape of AI governance. This piece examines the CCW Group of Governmental Experts meetings held in Geneva from August 31 to September 4 on lethal autonomous weapon systems, and their implications for military AI.

After more than a decade of discussions under the Convention on Certain Conventional Weapons (CCW), the 128 parties to the CCW agreed on a set of elements that could form the basis of a future international instrument on lethal autonomous weapon systems. The agreement was reached during the final 2026 session of the CCW Group of Governmental Experts on lethal autonomous weapon systems, held in Geneva from August 31 to September 4. The agreement followed three years of negotiations. While this agreement is not a treaty and does not constitute binding rules, the CCW Review Conference in November will show us more about what happens next.

To put the issue in a broader context, it should be noted that autonomous weapons are one part of military AI. AI is also used in intelligence analysis, targeting support, cyber operations, and military command systems. The UN is discussing these uses in different forums. The General Assembly now has a periodic process on AI in the military domain, while the Security Council has also discussed AI and international peace and security. Thus, the CCW negotiations deal with one part of a wider debate over military AI governance.

What did states agree on?

The agreed elements refer to lethal autonomous weapon systems as systems that can identify, select, and engage targets without a human operator carrying out those tasks. During the September 4 meeting, the Chair clarified that what matters is what the system is capable of doing. Also, he drew a distinction between a human choosing a specific target and a human defining a category of possible targets from which the system later selects one.

Moreover, states agreed that international humanitarian law applies to these weapons. Yet, the negotiations did not settle on how much human involvement is required. This became one of the main issues across the September 1, September 2, and September 3 meetings.

How much human control?

States still disagree over when humans need to make the decision and what can be left to the system.

For example, a commander may set the target type, area of operation, and time limits before activation. The system can then select and engage a target that meets those conditions. Other systems may require an operator to approve the strike. In both cases, a human makes the decision, but at different stages.

During the September 1 discussions, Singapore stated that constant human control may not be possible for systems such as air defense, where decisions may have to be made within seconds. Other delegations argued that humans still need to make the judgments required by international humanitarian law.

The U.S. policy takes a similar position. The Congressional Research Service notes that the Department of Defense policy requires “appropriate levels of human judgment” over the use of force, though not human approval for every individual engagement.

Furthermore, AI can shape targeting before the final decision is made. AI-enabled decision-support systems can analyze intelligence, rank possible targets, and suggest options to military decision-makers. In a recent study, Michael Horowitz and Erik Lin-Greenberg found that U.S. national security experts were more willing to support military action and had more confidence in the intelligence when the assessment was conducted by humans rather than AI. This shows that the use of AI in intelligence analysis can affect how decision-makers assess information before deciding whether to use force.

What happens next?

Human Rights Watch reports that 76 states support negotiations on legally binding rules, including 70 CCW parties. The United States and Russia remain opposed to a legally binding instrument.

During the September 3 discussions, several delegations said that the agreed elements form a basis for future work and would still need to be developed in later negotiations. The CCW Review Conference in November will make it clearer whether the group receives a new mandate to negotiate an international instrument.