Robin Geiss (UNIDIR) on the launch of a range of new initiatives from UNIDIR's Centre of Excellence on AI, Peace and Security - Press Conference

All right. We're delighted to be joined by Robin Geiss, who is, I mentioned, the director of the UN Institution on Disarmament Research, otherwise known as UNIDIR. So, Robin, welcome. You have the floor, then we'll take some questions.

Yes. Thank you very much, Stefan. Thank you all for joining us today. UNIDIR works on the key disarmament, arms control, and global security issues of our time, and AI is clearly one of them. So I'm here today to speak to you about artificial intelligence, a recent UNIDIR Center of Excellence on AI, Peace and Security that we launched in Geneva this summer, and a number of initiatives that we're going to unveil over the course of this week here in New York. So to start us off, I think we can say machine warfare is coming ever closer. And the question, put simply, that's in front of us is really, are we going to let machines target and kill human beings? Secretary General of the United Nations has warned repeatedly, we're dangerously close to crossing this moral red line. And let's make no mistake, artificial intelligence, autonomy are already embedded widely across the military domain. We find them in intelligence analysis, in target identification, drones, cyber attacks, information operations, and so on as we speak. Decision cycles are compressing. They're compressing to machine speed and the margin for human judgment is increasingly narrow, including in the nuclear domain. Opaque systems are increasing the risk of error, of miscalculation and escalation, unintended escalation. And the proliferation of AI capabilities means that the threshold, the entry point for very sophisticated operations is getting lower, potentially also allowing non-state actors to engage in sophisticated operations supported by AI. small number of countries and companies are shaping security outcomes in the domain of AI in ways that traditional arms control frameworks were not designed to address, and in ways that profoundly affect the security of all other states on the planet. Yet despite of all this, for a long time at least, there has been quite a reluctance to engage with this, perhaps the sharpest edge of AI's growing impact on the world, both within the UN and also beyond. Thankfully, this is beginning to change. There is a growing realization that we stand at a crossroads, that now is the time to act, including on military AI. And if this needs any proof, just look back to the recent high-level discussions here at the United Nations General Assembly that were so focused on artificial intelligence and some of the risks it poses. Now, to be clear, geopolitical competition and tension are serious obstacles to the way forward, of course. But concerns about AI's impact are already transcending geopolitical blocs and they're prompting action from major powers. Just last month, for example, we saw the United States propose an AI safety hotline with China in case of an AI incident that could threaten national security. And the Secretary General's repeated calls for stronger regulation of military AI and a legally binding treaty on lethal autonomous weapon systems increasingly find an echo across the international community. So it's against all of this, against the backdrop of all of this, that at UNIDIR, we recently launched the UNIDIR Center of Excellence on AI, Peace and Security. The center is a one of its kind, permanent, impartial platform within the United Nations that aims to present a coherent approach at the very issue of AI in international peace and security. This Thursday here in New York, together with Switzerland and Pakistan, we will launch the center's new forum, which will, once it's there, present the only multi-stakeholder body within the United Nations system that works on AI and international peace and security and in the military domain. Tackling this highly sensitive but increasingly crucial issue will, of course, require sustained engagement with actors from every relevant sector and all over the world, which is precisely what the forum and the center are here to do. Over the course of the week, three other center initiatives will be unveiled. The first addresses the role of industry and aims at developing a framework for responsible industry behavior. Decisions taken by companies on system design, training data, testing, support during deployment and post deployment, all these determine whether AI enabled military capabilities are ultimately reliable, predictable, and whether they can operate in line with international law. So it's a crucial stage, the design and development stage for which industry is largely responsible. And for that reason, UNIDIR is developing a framework of responsible industry behavior that will provide practical, voluntary guidelines grounded in international law and in the UN Guiding Principles on Business and Human Rights. This will serve to establish a common baseline for responsible conduct across the technology life cycle, which will go on to inform government procurement and public-private partnerships. The targeted consultations we're launching this week engage the full spectrum of industry stakeholders, major and emerging defense technology companies, software developers, dual-use service providers, subcontractors, and investors, all working alongside governments, civil society, academia, and expert communities. We're partnering with the Office of the High Commissioner for Human Rights on this to also boost coherence on this issue across the entire UN ecosystem. And since the framework's credibility will depend largely on the inclusiveness of its development, the process itself will be just as important as the final product. A second key initiative that we're launching this week here in New York addresses the issue of language. We find that states frequently use terms like autonomy, safety, or security to mean very different things. And this ambiguity then leaks international positions and approaches, and it gets into the way of finding agreement or building trust on the international level. This in mind, UNIDIR this Friday will launch a project to develop an AI security lexicon that will both clarify key concepts and also set out divergent understandings that remain to date. Ultimately, our hope is that this will help states to find common ground and to engage on an equal footing on this complex issue of AI in the military domain. The third initiative that we're launching this week addresses the risk of misuse of AI and AI capabilities by non-state armed groups. UNIDIR is working with UNITAR in Austria this week to convene a discussion on the non-proliferation of AI. focusing on misuse of AI capabilities by malicious actors such as terrorist groups and criminal networks. Emerging research provides concrete evidence of AI already being used by such actors to support operational planning, execution, and logistics. But despite the significance of the potential risks that are there, to date, this issue has received rather little attention. Existing non-proliferation regimes, expert networks, export control frameworks, they provide a foundation for dealing with such issues, but the challenge of AI clearly extends far beyond the traditional scope of these instruments. Our aim is to identify where gaps remain and to come up ideally with concrete measures that can be implemented at the national and international levels. while ensuring that efforts to prevent misuse do not limit access to the peaceful uses and benefits of artificial intelligence for all. Now, over the coming months, UNIDIR's Centre of Excellence on AI, Peace and Security will carry this important work forward, not least by helping states to develop national approaches to AI in security and defence. We will also start to build our support for Switzerland as it prepares for the Geneva AI Summit in June 2027. where the urgent issue of AI in the military domain will certainly no longer be allowed to linger in the shadows. The time has come to pool our strengths and confront AI in the military head on. This is what the center is here to do. Thank you all very much for your attention today, and I look forward to your questions.

Thank you very much. Edith, please.

Thank you very much on behalf of the United Nations Correspondents Association. My name is Edith Lederer from the Associated Press, and I have a bunch of follow-up questions. First, who's going to be funding all of this work? Is it from the UN or is it from private, private or government donations? And if so, do you have enough to get this project off the ground? Secondly, how narrow is the timeline between autonomous action and human control of military weapons and nuclear weapons.

How sorry, just to ask, how narrow is the timeline? How close are we to full autonomy? That is difficult to assess, and I think it's exactly that uncertainty that we cannot know with certainty how close we are to fuller autonomy is the issue we're trying to resolve here. One of the problems is, of course, that we're seeing AI is increasingly creeping in in so many different ways. It's not just a decision. Are we using AI for this decision? Yes or no? It is creeping into early warning systems. It is creeping into information analysis, reconnaissance, information gathering. And this creeping in of AI is, you know, a process. And it's quite difficult to say at which point we're actually really losing control. But we can say that this process is clearly on track to crowding out human decision making increasingly. And that's what we're concerned about. To the funding, as UNITID is a voluntary funded organization, I'm delighted to say that we have a very wide set of donors for the Institute, including all the P5 and countries from all of the regional groupings, which I find quite exceptional given the very sensitive issues that we're dealing with. We have a list of states that are funding the Institute. We have, so to say, startup funding that will get us through the initial five years. So we're comfortably funded for the center, and comfortably funded also means we can our autonomy in taking this research forward.

Do you, as a quick follow-up, do you have buy-in for this whole Institute and what it's doing from the P5, from the US, Russia, China, Britain, and France?

All of them are donors for different aspects within the Institute, but all of them are donors to INIDI, that's correct, yeah, as of today.

Great, thank you. Namo, and then we'll go.

Thank you very much. Nam Abdullah with Ruda Media Network. On two things. First, what do you think, what do you see as the strongest incentive for governments and militaries to hand decision making over to the AI when it comes to killing humans? This is just a speculation. Why would governments and militaries want to do that? And my second question is on the threat of biological terrorism. A few weeks ago, months ago, for the first time, an AI designed a complete genome, like a virus was created from scratch. And it was a good virus, but it could be a bad virus. Can you talk more about that? Thank you.

Yeah, on the biosecurity level, that is certainly, I think, a risk and concern that we need to take very seriously. As I said, we're trying to really cast a wide net engaging with so many different actors, whether it be at states or from the industry sector. And one concern that keeps reemerging and that's very much in the here of now, I mean, there's many different layers of risk we're discussing with regard to AI, but one of the more plausible, more imminent ones is clearly that very powerful language, large language models could be used to uncover things in the field of biosecurity that previously were simply not accessible. From there on, it's still a way to developing any weaponized kind of structures or systems, but nonetheless, this is a very serious concern. It is core focus of our research to addressing this issue and making sure that biological capabilities don't get into the wrong hands. On the loss of control and incentives there. I mean, the military never has any incentive to lose control over anything. So if anything, they want control. So that's not my priority concern. But we do see emerging arms racing dynamics. We do see that on the battlefield, the race around AI in the military is very much a race for speed, increasing the pace of battlefield operations. downsizing the sensor to shooter time, increasing the number of targets that can be engaged in any one day in military operations. This is already reaching unprecedented levels that we haven't seen before. So complexity and pace are increasing and there is a wider risk of losing control, not over individual weapon systems, but potentially if we're overdoing it over military processes. Plus we're seeing a transformation at a scale technological scale that simply we're not used to and we haven't seen at that speed. And I think this creates a number of uncertainties where we're simply not used to new types of battlefield behavior, new types of operations, swarms of uncrewed systems, and we're not so sure always where all the red lines are. And that's an important field of our work. That's kind of the different dimensions where I see that loss of control could be imminent. Another issue of loss of control is being discussed already for some years at the United Nations is with regard to lethal autonomous weapon systems. That's more specific systems, the tip of the iceberg, if you want. But AI in military decision making is a far broader and more complex issue in more general. Yes, please.

Hello, my name is Aynur. I represent Silk Way TV channel of Kazakhstan. I have two questions. One, follow up on the first initiative. You said the framework that will include, um, the, um, defense companies, software companies, other, uh, representatives of, of, of the stakeholders, have any of these industry leaders, um, uh, have really confirmed their participation in, in, in, in taking part of the discussions and being a part of, of this dialogue?

Yes, uh, some are already on board, we're launching consultations this week to bring in more, the idea that we have is that we want to cast a wide net, uh, on these issues, um, very diverse set of actors, like I said. And what we're certainly seeing is that compared to maybe five years ago, now the interest on the industry side is significantly increased. There's much more interest. We see many more industry players at our different conferences, we convene the different workshops that we run all year round. So yes, some have confirmed, some prefer not to be listed publicly, but we're in discussions with a very wide range of actors on this.

Could you disclose the names of those?

Microsoft, for example, is a donor to media and will be part of this process. Hitachi America, I just, sorry, I don't have the individual names, all of them present, but there is a list of companies. It's public, the ones that agree to be listed anyway.

And my second question is, despite all of these concerns regarding nuclear disarmament, we can see that all nine nuclear-armed states continue to expand and to modernize their arsenals, and it's make nuclear disarmament topic increasingly difficult to achieve. And the numerous experts are warning of a high risk of nuclear war. So my question here is more general. What mechanisms do you believe could effectively bring nuclear powers together for a dialogue on nuclear disarmament? Because at the moment, they show unwillingness of addressing this topic.

Yeah, there is a range of fora to pursue We had the NPT review conference earlier this year, of course, which was a milestone event in and of itself that is one of the core fora to discuss these issues continuously. Of course, we all know that we are in a very difficult geopolitical environment. For us, this means here too, with regard to nuclear weapons, we're trying to be creative about formats to convene at the expert level as many dialogues as possible to understand the very fast changing landscape, because what's changing in AI and new technologies, what's changing in outer space, all these different dimensions converge and they all have nuclear implications as well. So there is significant uncertainty. Maybe to give just one example with AI, if with increasing sensors and more autonomous systems that roam the oceans, at some point it might be possible to track and identify submarines in a more coherent way than is currently possible. That could compromise second strike logics and current deterrent logics. So there is a lot to unpack in this area, and this is what we're working on at UNIDIR.

Thank you very much. Edie, one last one.

One last question. The prospect for any kind of international cooperation on the military use of artificial intelligence has been exceedingly limited in the past. What do you attribute the growing interest in this field too.

Yeah, well, just, you're right, and I agree with that assessment, but of course, now we're seeing these technologies actually applied at scale. And as we're going to see more advances in the field of robotics in the coming years, all of this will become significantly more real than so far it has been and has been perceived. And I think this is, after all, what is prompting states to address this issue in a more structured way. And in times of geopolitical competition, ongoing conflicts is, of course, a high-stakes issue. And we have to manage our expectations. But at the same time, we're seeing some of the very grave risks unfold in front of our eyes. There really is no time to lose on these issues. And so we hope we can do our share to move the international community towards more progress in this important field.

Great. On that note, Robin, thank you so much. Thank you very much. And thanks to all of you.