With the increasing use of AI on the battlefield, particularly in the wars in Gaza, Iran, and Ukraine, how states address its governance is increasingly under the microscope. Efforts towards effective multilateral governance have the potential to specify what is – and is not – permissible under international law with regard to AI in the military domain (AIMD). Not all states are eager to see that outcome. Others, however, fear that an absence of meaningful guardrails will undermine compliance with international law and human dignity in armed conflicts.
A series of different intergovernmental fora have been established to attempt to address some of these issues, which are introduced below. Some states view these various diplomatic initiatives on military AI and lethal autonomous weapon systems (LAWS) as failing to produce meaningful outcomes, whilst the technology continues to develop at a rapid pace. Many states want to see new legally binding instruments, starting with LAWS, in which they hope to ban autonomous weapon systems and establish common standards for the development and use of AI systems for defense and military purposes. Those states with strong militaries and advanced AI capabilities, observing how successful the use of AI on the battlefield can be, are more cautious or in some cases outright opposed to new law, resistant to potential constraints on their own military use of AI. Multilateral dialogue on military AI finds itself at a crossroad, notably as current efforts tend to be overlapping yet dispersed, with several different potential routes: either continuing with the current fragmented dialogues or pulling the multilateral process under one comprehensive umbrella.
After many years of multiple diplomatic initiatives on military AI and LAWS, in December 2025, United Nations General Assembly (UNGA) resolution 80/58 mandated informal exchanges on AIMD for states to discuss the issue in a U.N. context and to set out their views on next steps, possibly including the establishment of a new, consolidated U.N. process through the UNGA. The first informal exchanges were held at the United Nations in Geneva from June 15-17, 2026, and point to a possible more permanent way forward. Specifically, they have the potential to streamline and universalize intergovernmental dialogue under a new, consolidated forum on global governance and international norms on military AI and related issues. The informal exchanges were followed by the Global Conference on AI, Security and Ethics on June 18 and 19, organized by the United Nations Institute for Disarmament Research (UNIDIR).
What emerges from this discussion in Geneva is a desire to create a new architecture for discussions on AIMD that is genuinely universal and anchored in the U.N. However, given the patchwork of existing initiatives and great powers’ reluctance to engage in joint normative enterprises, it will not be straightforward to achieve.
Current Processes: A Fragmented Playing Board
Military AI and LAWS are the object of several multilateral processes. The Group of Governmental Experts (GGE) on LAWS was established in 2016 by the Convention on Certain Conventional Weapons (CCW).The GGE aims to develop guidelines on LAWS, potentially including their prohibition. Only the 128 states that are high contracting parties to the CCW have voting power, although all states and non-governmental actors can attend the GGE. The GGE is due to conclude its work on a set of elements for an instrument in September. It is not yet clear whether its mandate will be extended for further work, and if so, in what scope; these decisions will be made at the conclusion of the Seventh Review Conference of the CCW due to take place on November 16-20, 2026.
In addition, the Netherlands created a standalone process on Responsible AI in the Military Domain (REAIM), that has led to three summits: in the Hague (2023), Seoul (2024), and A Coruña (2026). In so doing, the Netherlands was aiming to expand the discussions beyond LAWS, establish a multistakeholder process, and inject new normative momentum among states. Although open to all states (except for Russia, which was not invited due to its invasion of Ukraine), not all states attended the summits. While some anticipated that the Third REAIM Summit in A Coruña in February 2026 would be the last REAIM meeting, Kenya used the AIMD meetings in Geneva to announce that it would host the Fourth REAIM Summit in Nairobi in April 2027. This shows that the REAIM co-sponsors still see value in these summits’ power to convene a range of stakeholders, including importantly the private sector, to discuss military AI. Although not as supportive as it has been previously, it is worth noting that the United States has not rejected the REAIM process in the same way as it has rejected a U.N. role for AI governance writ large. The hosting of the next Summit in Africa will be an opportunity to show that REAIM remains a relevant forum and engages the countries of the global south.
The topic has also been raised periodically at the UNGA, but no formal U.N. process has been created out of that forum. Similarly, the United States had led a process on a joint Political Declaration on Responsible Military Use of Artificial Intelligence and Autonomy, yet has suspended its efforts despite considerable traction among like-minded states.
The result of these many streams is a fragmented multilateral approach to any kind of governance of AIMD, be it legally binding or voluntary. This is a significant gap in the effort to determine whether, and how, to govern the use of AI in the military domain that risks confusion, a loss of focus, a scattering of efforts, and ultimately a loss of momentum towards joint progress.
A New Multilateral Process
The informal exchanges in Geneva on AIMD show that a new multilateral process can address this gap. The discussions involved states from across the globe as well as researchers and industry from various countries. While they covered issues and challenges typical to other diplomatic initiatives, such as REAIM and the GGE on LAWS, the essence of the meeting concerned multilateral processes.
Most states expressed a desire for a new multilateral and universal process at the U.N. For many, the obvious choice was an open-ended working group (OEWG). Several states said that careful thought would need to be given to the group’s mandate. Although it would be on the broader theme of AIMD, in order to be effective, these states emphasized that a new OEWG should focus on specific topics such as confidence building measures or how to design responsible systems.
The meetings in Geneva also involved a certain amount of institution-building. On June 17, UNIDIR launched its Centre of Excellence on AI, Peace and Security. The new centre will serve as a permanent platform dedicated to advancing AI governance in the peace and security domain. It will bring together research, dialogue and capacity-building to support more effective and inclusive approaches to the challenges AI presents. Although this cannot substitute for an intergovernmental process, it does provide a stable home for discussion, exploration, and institutional knowledge on AIMD in Geneva and can provide a useful platform for exchanges before and after the establishment of a new multilateral process.
However, the sponsors of the AIMD resolution, the Netherlands and the Republic of Korea, together with their core group of supporting countries, have a timing and overlap issue to think through. Many states will want to know the future of the GGE on LAWS to be decided at the November CCW Review Conference, before committing to establishing a new AIMD process.
The situation gets more complicated when you factor in the Austrian sponsored UNGA resolution on LAWS adopted in December 2025, which calls on the CCW states to complete their work on elements for an instrument, but falls short of creating a new multilateral process. Austria too will be looking at what happens to the GGE on LAWS to determine whether to establish a new process in the UNGA.
To add to the complexity, the CCW’s decision about the future of the GGE on LAWS will come after the UNGA First Committee on Disarmament and International Security in October 2026, where in theory states could formally decide to launch a new multilateral process on AIMD. The First Committee has no role in determining the fate of the GGE, and cannot take on the issue of its extension in that forum. States therefore face a dilemma: either launch new processes that may or may not support progress in the GGE during the First Committee meeting, or wait for the GGE’s results and take it from there, but risk losing time and momentum.
With all of this taken together, the sponsors of these resolutions need to think about having too many overlapping processes that cover similar ground, which is unlikely to aid reaching agreement on anything approaching a new legally binding instrument. They should also be mindful of process fatigue. There have been a lot of U.N. working groups recently, on topics such as biological weapons, cyber, and outer space. It is already difficult for states of all sizes to engage properly with these processes, and it is always a challenge to fit them into the already crowded disarmament calendar.
Given that the same set of states that introduced the AIMD resolution are also the ones that are continuing with the REAIM summits, one could argue there is a tension here between parallel processes, one formally universal and the other not. These states would likely respond that this tension is manageable, as the REAIM summits seek to bring together stakeholders from different sectors and would complement the work of any universal process. And yet, the hunger for a consolidated multilateral process persists.
Differing Approaches from the Big Three
The complexities articulated above are further textured by differing approaches between three big players in AIMD: the United States, China, and Russia. As a result, including them in the short-term – or, at least not excluding them in the long-term – requires a calibrated approach for deciding the next multilateral steps. Any U.N. process needs to take the positions of these three states into account. At this stage, however, it is difficult to say if any such process can lead to their convergence.
When U.N. member states voted on the AIMD resolution in October 2025, the United States voted against its adoption. It was not the only country to do so. However, it also chose not to attend the informal exchanges that took place in June. It was the most notable absentee from those exchanges. Its absence was consistent with its argument at the adoption of the resolution that determining the future of AI at the U.N. is a gross violation of national sovereignty. Earlier this year, citing similar reasons, the United States opposed the inclusion of an agenda item at the U.N. Disarmament Commission (UNDC) on emerging technologies. Many states had wanted to use the UNDC to agree on recommendations related to emerging technologies and international security. The U.S. position forced the body to meet only informally to discuss working methods.
It is almost certain that the current U.S. administration would not engage directly on any U.N. process on AIMD, which would weaken claims that such a process was, in fact, universal. However, many would argue that this should not be a reason to hold back from establishing a new process (which the U.S. could always engage with later). The measure of universality, after all, is not whether all states do in fact engage with a consolidated multilateral process, but whether they can if they so choose.
Russia also voted against the AIMD resolution. However, unlike the United States, it was present at the informal exchanges and submitted a working paper. Echoing that working paper, Russia gave a statement indicating that it did not object to establishing a multilateral dialogue on AIMD, but that any such dialogue must be based on consensus, inclusivity, and, with a nod to the absent United States, engagement from all the big powers. They criticized the REAIM process, to which they had not been invited, as being “destructive” and an attempt by a narrow group of countries to impose their ready-made solutions on others.
Perhaps mischievously, Russia proposed the UNDC as the forum to host AIMD discussions (mischievous, because the U.S. position prevented it from meeting formally this year). Russia argued that the universal membership, deliberative mandate, and the specialized nature of the work meant that the UNDC was ideally suited for such dialogue. They stressed its experience in developing recommendations and indicated their view that it could reach agreement on practical guiding principles on the use of AI for military purposes.
In choosing to promote the UNDC as a venue for the AIMD discussions, Russia is well aware that this would likely be opposed by the United States. Even if it were to be accepted, it is important to note that the UNDC operates by consensus, not by the UNGA voting rules that only require a simple majority (Article 18, U.N. Charter), meaning any one state could block anything they do not appreciate. The consensus rule is often blamed for the lack of progress on arms control and disarmament; however, for many states it remains a vital means of protecting national security because they can easily prevent any outcomes that go against their interest. This explains why many major military states want the LAWS discussions to remain in the CCW (which is also consensus-based), rather than go to the UNGA.
As with the United States, the absence of Russia from a new U.N. Process would weaken its claim to be universal (with the same caveats noted above). However, unlike the United States under the current administration, Russia is more likely to engage with the process, even if it is simply to ensure that it does not do anything it does not like.
Unlike Russia and the United States, China voted in favor of the AIMD resolution. China also attended the informal exchanges and was supportive of establishing a new U.N. process. Like Russia, China did not miss the opportunity to criticize the United States whilst they were not in the room, accusing the United States of being weak on oversight over their AI companies. Although China’s public position is in favor of a new process and a legally binding instrument on LAWS, as a major military with advanced AI capabilities it will likely be cautious about moving too quickly with anything that would constrain its current military advantages.
Conclusion
There is now a clear direction of travel towards an inclusive, multilateral, U.N.-mandated process on AIMD. A resolution to create a new process in Geneva, such as an OEWG, would likely get broad support among states. However, the interplay of any new process with the current ones will need to be thought through, meaning that we may not see this universal process established in the near term.
Further discussion will also be needed about what topics a new U.N. process prioritizes, the role of civil society, and the rules of procedure, particularly related to voting (such as whether the decision-making threshold will be a simple majority versus consensus-based decision making). If the sponsors of this new process offered the United States and Russia a framework that required consensus for any outcome, that could make them more open to engaging with it, as they would know they could ultimately prevent consensus should red lines be crossed.
Last month’s informal exchanges, conference, and the new Centre of Excellence also showed that Geneva may be becoming the permanent venue where all relevant stakeholders can come together to discuss AIMD. The organizers of the next REAIM Summit will need to think about what that process’s added value can be. Greater access for global south countries to the stakeholders developing the technology would be a good place to start.
Given the different processes and fora for discussing AIMD, there is never going to be a perfect means of bringing the issue into one place. However, a universal, multilateral process on AIMD can function as a center of gravity. An OEWG tasked to focus on specific aspects of armed forces’ use of AI and to produce recommendations seems the most integrative and realistic way forward.
