AI decision-support systems are often seen as just battlefield tools, but they are so much more

Terry Gerton You have published a paper around uses of military AI and coalition frameworks, but let me just back up for a minute. I think when most people think about military AI, they’re thinking about targeting or autonomous weapon systems. Where in the coalition space do operations most break down today and how can AI perhaps help mitigate some of those?

Sophie Mayo Yeah, absolutely. I mean, I agree that the most well-known use case of AI in the military at the moment is targeting due to recent operations in Ukraine and Iran. We do address this theme in the report, but there are actually other coalition challenges that my co-authors, Amelia Probasco and Lauren Kahn and I address in this CSET paper, Coalition Operations, the untapped role of AI in military decision-making. And we argue that these are ripe to be addressed with AI decision support systems or AI DSS for short. So some examples, I mean, first of all, the challenge of establishing a coalition operating picture. So this is really about building and maintaining a comprehensive picture of all military and civilian assets across a geographic area. And this is a really difficult challenge. It’s unhelped by the fact that coalition partners often lack the systems to ingest, to display, and to manipulate op data and information sharing policies. And that’s the really administrative piece which can delay the flow of information to allies. And our assessment here, just very briefly, I’m sure we’ll get into it further, is that these AI decision support systems can support in two ways. So. Common operating picture data sharing management to flexibly integrate data and algorithms, and also helping to manage complex information sharing challenges. The second meaty challenge that people may not have heard much about that we look at is implementing information sharing policies. For example, things like the foreign disclosure process. So in international coalitions, government policies, domestic laws and international agreements. All shape the releasability of certain pieces of information to different allies and partners. In some alliances, more mature alliances like NATO, these disclosure rules are largely settled. But in other cases, information disclosure requests have to navigate and weave through changeable and often competing interests, frankly, of intelligence agencies, operational warfighters, domestic politics as well as law. And this is the complex bureaucratic piece that again, these AI decision support systems can help with very briefly, things like implementing straightforward guidance, managing policy changes and supporting interpretation of different information sharing policy disclosure guidance. The final challenge, which again is perhaps a little counter-intuitive or surprising, is bridging language differences. So language and communication, fundamental factors for coalition operations for their success. And yet, through our research, we found shockingly numerous examples of confusion at best or near fatal accidents in the field due to language difficulties. Things like the mispronunciation, for example, of specific towns. And again, this is an area where AI decision support systems can help while the translation technology, admittedly, is far from perfect at present, the highly structured and standardized and jargon-laden nature of military technology is pretty well to these AI systems.

Terry Gerton Sophie Mayo is a graduate research assistant at the Center for Security and Emerging Technology. Sophie, it strikes me that some of those issues you just ran through are problems within a country’s military, much less across military. So I can only imagine the additional complexity in coalition operations. But there’s one piece I especially want to bring out because every country comes to coalition operations with its own rules of engagement, Right? Making sure that it’s Forces follow its laws and requirements. How much more complex is that in a coalition operation? How do leaders manage that and how can AI decision support systems help keep everybody on the right side of the law?

Sophie Mayo Yeah. So, I mean, one, there are many sort of flavors of this, but one specific example is this thing called a national caveat. So these really set the terms of reference and describe the mission, command, relationship, organization, logistics. It goes on and on accounting procedures, coordination and liaison of military units and personnel that may be detailed to a particular multinational force. And if I may just bring out with an example what this can look like in practice and what it has looked like in the past in the pre-AI era. So in Afghanistan, I mean, the NATO-led International Security Assistance Force was at one point 51 nations in one coalition. And so commanders and their staff were forced to sort of pass through really what has described as enormous spreadsheets to figure out. And manage and track these national caveats and sort out which forces and it could be detailed where and at what time with what weaponry. And on occasion, you know, this administrative heavy task even led to near accidents again in the field. The point, of course, in highlighting these examples is to highlight the art of the possible with AI decision support systems. So these systems can be used to create a centralized repository of national caveat. And data on available weaponry and platforms, meaning planners can spend less time searching for both these things in the enormous spreadsheets of conflicts past, and more time exercising judgment on how to allocate and assign forces.

Terry Gerton Sophie, in training operations, you can imagine where you have time to test and evaluate the kinds of AI tools that you’re talking about here. You can imagine that. But in combat operations, in real time, when you are in life or death situations, how would combat leaders use these tools? How would it impact how they are executing missions, perhaps under fire?

Sophie Mayo Well, I mean, there’s, again, a lot of examples that we could look at. There are a lot back office functions that I’ve mentioned that can be addressed very effectively to support more effective coalition operations. Let’s take one example, creating a common operating picture. Historically, this has been really, really challenging to integrate, flexibly integrate different coalition data sources into one picture. These systems can help with data sharing management, which makes a commander’s job and makes their staff’s job much less full of friction. So they can flexibly, these systems can flexibly integrate data and algorithms, for example, flexibly adding a new drone sensor feed. And the net effect of this particular example is that… Unlike in operations in the past, coalition data from nation-states, not-for-profits, even commercial technology providers can be added more rapidly to enhance an operational picture.

Terry Gerton So, as we look at these, you’ve got some examples in your paper of how NATO really seems to have moved fairly quickly to implement some of these decision support systems. What conditions were already in place there that allowed them to move so quickly and What kind of impact have they seen from the deployment?

Sophie Mayo Yeah, I mean, NATO really benefited from having infrastructure policy frameworks in place and also political will that turned out to be preconditions for the successful adoption of NATO MAVEN, which is their coalition AI decision support system. As you’ve hinted at, these factors contributed to NATO MAVEN becoming the fastest acquisition in NATO’s history. It’s pretty astounding. I mean, six months is all it took from outlining the initial requirement to acquiring the actual system, which happened in March 2025. And some of the outside of targeting uplift, which again, and as you mentioned at the outside is perhaps one of the most famous examples of what these systems can provide, they’ve actually been able to spend, to save rather a lot of money on training exercises. So millions of dollars—multiple millions of dollars in some cases on training exercises. Which ultimately at the end of the day also saves the taxpayer money.

Terry Gerton So what lessons do you want other alliances to learn from how NATO progressed through this process and what AI decision support systems can do to help?

Sophie Mayo Yeah, I mean, with reference to the NATO case specifically, I think there are two interesting lessons to point out. First is that allied confidence can be built through, and should be built, through demonstrations, exercises, and transparency, and Task Force Maven’s outreach, I think, offers a replicable model. Ultimately, the more candid allies can be about discussion of AI advantages and incidents, the better and the more comfortable they’ll be adopting these systems. The second is that given the success of task force, force maintenance, it may be prudent for the Pentagon, for the Chief Digital and Artificial Intelligence Office to treat this particular example as an operational asset and a learning laboratory to replicate and accelerate adoption with other partners in the Pacific. I’m Australian, so I’ll use Orcus Pillar 2 as one example of a possible coordination forum, which I think makes sense as the next point of call.