The drumbeats of war are getting louder — not only in Europe and the Middle East, but also in the Arctic, the Horn of Africa, the Taiwan Strait and elsewhere. The UN Secretary-General reports that global military spending could rise to between US$4.7 trillion and US$6.6 trillion by 2035. The latter level would be more than double the 2024 outlays — and five times higher than when the Cold War ended.
By contrast, far fewer resources and attention are flowing into harnessing new tech to prevent bloodshed. Or to simply reduce loss of life and assets amid an ongoing crisis. Still, a growing number of predictive artificial intelligence (AI) tools can credibly claim to forecast attacks or pre-identify hotspots.
In June 2026, risk consultancy Verisk Maplecroft provided Wall Street firms with access to new machine-learning models to gauge any given country’s likelihood of descending into war over the next 12 months. The products weren’t available before bombs started dropping on Iran on the last day of February 2026. Had they been, retroactive testing suggests they would have given that outcome a 66 percent probability weeks in advance.
Prior to that, Conflict Forecast — a conflict-prevention organization run by a network of researchers and data scientists — correctly foresaw both the Wagner Group’s revolt against Moscow in June 2023 and US intervention in Venezuela in January 2026.
But complications linger. Models’ performances are impressive, if nowhere near flawless. And then there’s the question of what to actually do with this information.
“We ultimately are kept in the dark about how policymakers use our predictions,” Christopher Rauh, a University of Cambridge professor and founder and co-director of Conflict Forecast, wrote over email. “We know that they use them as a source of information, but we do not know whether they actually act on them. And when they do, neither do we know how.”
This resonates with Branka Panic, founder and executive director of AI for Peace. “The field has made remarkable strides in data availability and forecasting accuracy, yet the warning-to-response gap remains the central unresolved challenge,” she told me.
The Future Comes into Focus
Using digital tools to predict eruptions of violence isn’t new. Indeed, applying algorithmic rigour to conflict analysis dates back to the 1990s. Only recently has the technology caught up to the concept.
Amid a crescendo of Taliban attacks in Afghanistan in late 2019, an overstretched US military became desperate to optimize resources in a wildly expensive and deeply unpopular deployment. In response, it developed an AI-enabled program called Raven Sentry to predict assaults on Afghan cities.
A handful of army information technology officers worked with private-sector firms to build a bespoke AI model that digested four decades’ worth of reports on Afghan insurgents’ behaviour, movements and tactics against occupying forces — beginning with Soviet invaders in the 1980s. This information was combined with various environmental and social factors known to make executing raids easier. For example, weather forecasts calling for clear nighttime skies. Or the timing of religious holidays — when security forces tend to let their guard down.
During a year of testing, Raven Sentry accurately predicted more than 41 insurgent attacks on Afghan government facilities. In most cases, warnings came with at least 48 hours’ advance notice. Although US forces’ abrupt withdrawal from Afghanistan in August 2021 meant the model was fully operational for less than a year.
More recently, the United Nations Development Programme (UNDP) has partnered with tech firm CulturePulse to game out pathways to peace in Northern Ireland, the Balkans and South Sudan. The company specializes in creating digital twins of target-market audiences for consumer brand development — expertise repurposed to construct simulations involving millions of AI agents trained to represent opposing sides in high-risk situations.
CulturePulse’s “artificial societies” factor in granular data points on not only cultural identity and political ideology but also prevailing norms and beliefs and individual psychology. Simulations are then run repeatedly, toggling conditions under which conflict and cooperation might increase or decrease. This process has deftly pinpointed subtle drivers of conflict and cooperation in each region.
For example, the model found that in Northern Ireland, media-driven public anxiety was aggravating conflict. In the Balkans, expressions of negativity that fed into contestation over dominant cultural beliefs. In South Sudan, race and ethnicity concerns. On the flip side, cooperation in Ireland primarily took the form of constructive responses to negative rhetoric about race and ethnicity. In both the Balkans and Sudan, it was best generated by addressing personal household-level concerns.
“Such tools do not tell us which policies to implement, but they do inform such decisions by illuminating the actual drivers of conflict or cooperation in each region,” reads a UNDP background paper detailing the project.
Elsewhere, AI technologies are being used in Africa to funnel social media content and citizen-generated data into the African Union’s Continental Early Warning System.
Geospatial analysis firm Vantor — which counts more than 60 government clients — last year unveiled a first-of-its-kind automated threat intelligence tool that continuously synthesizes information from multiple satellites. The company says it can help its customers track hundreds of risk variables across the globe simultaneously, comparing them to historical imaging data to predict threats before they emerge. This technology covers everything from rivals’ military base activity to illegal fishing operations. Updated data sets and analysis are sent to decision makers every four to six hours.
All told, the promise of these technologies appears very real. The bigger question, though, is whether officials are ready for them.
Uncomfortable Choices Upfront
As with all AI applications, these systems are only as good as the human judgment steering them.
For instance, the UNDP reports that CulturePulse’s techniques have great value. However, the agency also warns that the simulations’ findings will only support the peacebuilding process if key stakeholders are involved and feel accurately reflected in the models themselves.
Forecasting tools also tend to generalize. “Early warning systems based on AI paint with a relatively broad brush,” wrote Rauh in an email.. “An early warning system is going to produce false positives — situations where the alarm rings, but nothing happens.”
This points to a giant dilemma for policy makers. Adopting conflict-prediction tools can provide crucial insight in time to prevent a crisis from flaring or spiralling out of control — but only if officials rapidly deploy resources upfront. And, more importantly, that commitment must endure even if predicted incidents fail to materialize from time to time.
“This sort of preventative mindset can have financial and political costs, but the benefits should outweigh them,” wrote Rauh. Given the damaging legacies of war, he argued, waiting for absolute certainty to act basically means waiting until violence has broken out before responding. At that point, it’s too late to avoid lasting fallout.
“Better forecasts do not, on their own, generate timely or context-sensitive action,” warned Panic. The problem is no longer access to information, she added. Instead, it’s converting information into hard decisions about where attention and resources should go. Action might include mobilizing military deterrence or launching a diplomatic blitz, but also pre-positioning humanitarian aid or getting logistics in place for civilian evacuations. “No tool pre-empts a crisis on its own,” Panic explained. “Closing that gap requires as much investment in human judgment, institutional relationships and decision-making mechanisms as in the models themselves.”
Another paradox: predictive conflict tools might ultimately be repurposed for target acquisition. There may already be a dangerous precedent for this.
The US military’s Project Maven initiative began in 2017 as an effort to reduce lethal targeting mistakes by American forces in the heat of battle. The goal was to blend AI, computer vision and dozens of other data points into detailed risk assessments to better distinguish between civilians and combatants. The program received only tepid support at first. However, once it dawned on officials how the same tools could bolster offensive capabilities by vastly expediting targeting decisions, its budget and importance both skyrocketed.
Fast forward to 2026 and Project Maven’s main platform — the Maven Smart System — enabled US and Israeli forces to jointly hit a combined 4,000 targets in the first four days of fighting in Iran. What was once a niche project has morphed into a pillar of the modern American war machine.
In an increasingly hostile world, nations now have access to cutting-edge tools for peace. But even these technologies underscore a timeless truth: achieving stability remains a matter of political will above all else.
