How NATO can integrate AI to prevail in future algorithmic warfare

A seemingly unbelievable fact: according to experts interviewed by NATO, the most vulnerable link in AI-powered warfare is not technology, but human beings. It is not chips, underwater cables, or algorithms that are the real weak point: it is our ability to manage errors, cognitive biases, and disinformation. And the most surprising thing is that, despite all the hype about “autonomous weapons” and “AI changing the face of warfare,” the risk of nuclear escalation does not depend on the use of AI itself, but on the effects on the ground and on human decisions made under pressure. The central argument of this report is that AI does not create fundamentally new vulnerabilities compared to the risks already present in cyberspace. What is changing is what is at stake: speed, scale, autonomy, and a new level of complexity that can amplify errors and misunderstandings. Consider how NATO is moving forward: it is not focusing on an “AI superweapon,” but rather on the ability to integrate artificial intelligence into every layer of its digital infrastructure, making it a sort of connective tissue for decision-making, logistics, command, and control. This is where Dominika Kunertová, the researcher who led the study with support from NATO and the Atlantic Council, comes into the picture. During a workshop in Washington – held strictly behind closed doors – crises were simulated in which autonomous drones filled the skies over the Baltic Sea. The responses most favored by the experts? Not missiles or cyberattacks, but diplomacy and electronic warfare: jamming, sensor blinding, and electromagnetic spectrum warfare. The use of nuclear electromagnetic pulse weapons, even as a mere threat, was deemed too risky and politically unsustainable. One key point to remember: today, the true strength of AI in the military sphere lies on three fronts, known as the “AI triad”: data, algorithms, and computing power. Each of these can be attacked—from data centers to manipulated algorithms to sabotaged chips in the supply chain. But what is the most effective defense? Redundancy: backup systems, transmissions over secondary channels, and realistic drills in environments where communications fail. And, above all, training: true “AI literacy” is not about knowing how to program, but about understanding the limitations, risks, and potential of AI, so as not to fall victim to the “allure of the machine.” A striking scene: In Ukraine, the drones used by the military share information as if they were Uber, assigning targets in real time to the drone that is closest and most ready. Everything works as long as the network holds up—but all it takes is an electromagnetic blackout, and that superiority is wiped out. And here, an uncomfortable truth emerges: the true frontier is not just cyberspace, but the realm of the electromagnetic spectrum. If you lose control of frequencies, communications, and GPS, you lose the war of algorithms. An aspect that is often overlooked: AI is not a magic wand, but a tool that works only if the underlying structure—clean data, secure hardware, trained operators—is robust. And beware of the risk of “AI hype”: the anxiety of being left behind can lead countries and generals to overestimate the actual capabilities of AI, creating more instability than the technologies themselves. The sentence that sums it all up? AI does not change the nature of warfare, but it does accelerate its pace and uncertainty. If this perspective has helped you see digital warfare in a new light, you can mark I'm In on Lara Notes: it's not a 'like'; it's your way of saying that this idea is now yours. If, in a few days’ time, you find yourself telling someone that the real vulnerability in AI warfare is humans—or that jamming matters more than killer drones—on Lara Notes, you can tag the person you were with using Shared Offline. That way, they'll know that conversation mattered. This Note is taken from an Atlantic Council report: you’ve saved yourself over an hour of reading.