Topic 39: What is Defense AI?

_{Was this email forwarded to you? Forward it also to a friend or a colleague! Sign up}

From the editor: I asked my friend, Dr. Larysa Visengeriyeva, partner at European Defense Tech Hub and co-founder of Women in Defense Tech, to write an article about how AI is used in defense tech. We often talk about AI as a technology, but it’s just as important to show how it’s actually being used. And what could be more tangible – and unsettling – than its military applications?

Introduction

Artificial intelligence is reshaping modern defense. Think autonomous vehicles in land, sea, and air; sensor fusion that stitches a live map from terabytes of raw data; logistics routes planned on the fly; war-game trainers that learn from every drill; and cyber systems that catch intrusions before they spark. Together, these AI layers compress decision cycles, sharpen situational awareness, and boost operational efficiency – that’s Defense AI. The ongoing war in Ukraine has accelerated these developments, turning the country into a testing ground for cutting-edge military AI. Ukrainian officials have actively partnered with Western tech firms to make Ukraine “the world’s tech R&D lab” for defense, deploying experimental AI solutions on the battlefield in ways that NATO countries are only beginning to explore [2]. At the same time, NATO and its allies have recognized AI as critical to future security. In 2021, the Alliance adopted its first AI strategy, followed by a revised one in 2024, which set priorities for responsible use, interoperability of AI systems, and combining AI with other emerging technologies [5]. 

In this article, we will focus on five key defense tech domains:

• Autonomous drones and multi-agent systems 

• Battlefield decision-making and situational awareness 

• Logistics and maintenance 

• Training and cognitive readiness

• Cybersecurity 

We will highlight both deployed solutions and ongoing R&D by showing Ukraine’s wartime innovations and NATO’s initiatives.

Autonomous Drones and Multi-Agent Systems

Autonomous drones and multi-agent “swarm” systems are at the forefront of AI in defense. Militaries are increasingly exploring coordinated swarms of unmanned aerial and naval vehicles for reconnaissance, targeting, and strike missions with minimal operator input.

In theory, networked swarms act as force multipliers – covering wide areas or saturating defenses in ways a single UAV cannot. The U.S. and its allies have tested swarming for over a decade, but true operational swarms remain in development due to technical hurdles [6]. Defense firms like Anduril Industries have showcased AI-enabled drone swarms that can be launched from land, sea, or air and coordinate ISR or strike operations with limited oversight. These drones maintain formation, avoid collisions, and divide tasks using onboard AI. NATO strategists see interoperable swarms as a cost-effective way to project force across allied forces without relying on billion-dollar manned systems. In early 2025, Sweden unveiled a swarm of 10 quadcopters that provided live battlefield feeds to troops – essentially acting as “flying eyes and ears.” A single operator commands the swarm via tablet, while an AI handles coordination and tasking.

From the start of the Russian invasion, Ukraine leaned heavily on drones – from Bayraktar TB2s to modified quadcopters dropping munitions. While most early operations were human-controlled, by 2024 Ukraine began deploying thousands of increasingly autonomous drones [3]. These platforms use onboard AI for navigation and target recognition.

One key innovation is the “ZIR” (ukrainian: “eyesight”) module – a small AI-powered kit that gives drones automated navigation and targeting. Trained on a vast imagery dataset, ZIR can detect and lock onto enemy vehicles from up to 1 km away, even tracking moving targets at 64 km/h. It adds autonomy via ArduPilot, enabling drones to fly pre-planned routes and return to base without GPS using optical terrain matching. In trials, AI-enabled drones raised hit probability from 10–20% to 70–80%.

It’s important to note that Ukrainian doctrine still keeps a human in the loop – operators approve targets, and AI takes over for the final attack run. Fully autonomous swarms, where drones make collective decisions in real time, haven’t yet been deployed. But small-scale experiments are underway. Ukrainian engineers view this as the next step – one that will require stronger AI coordination and communications. For now, the goal is mass deployment of semi-autonomous drones to conserve manpower and overwhelm Russian forces with “drones, lots and lots of them,” as one Russian soldier grimly described.

Generally, AI enhances specific military functions, such as: 

• Drone footage analysis 

• Target recognition and tracking

• Autonomous navigation (including last-mile navigation) 

• Sound and text analysis for intelligence extraction

Within NATO itself, there is a parallel push to exploit autonomous systems. The Alliance’s Future Combat Air programs, for example, envision “loyal wingman” drones, which are semi-autonomous UAVs that fly alongside manned fighters to extend sensor range or carry extra weapons. 

Prototypes like the UK’s Mosquito program and the U.S. Skyborg project have tested AI pilots that can navigate, avoid threats, and engage targets under human supervision. On land, NATO armies are experimenting with autonomous ground vehicles for scouting and logistics, and at sea, autonomous naval vessels (surface and underwater) are being trialed for mine-hunting and patrol roles. NATO exercises such as the 2023 ROBOTICS trials and the Dynamic Messenger drills have explicitly tested how mixed swarms of air, land, and sea robots might work together in contested environments [8]. 

The Alliance recognizes that to keep a technological edge, it must deploy swarming and autonomous systems, not just counter them [6]. That said, NATO’s approach is generally cautious, emphasizing safety, reliability, and ethical use. Autonomous weapons remain a sensitive topic and allies insist on meaningful human control over the use of force, in line with NATO’s AI principles [1] [3]. 

As technology matures, both Ukraine and NATO expect autonomous multi-agent systems to take on increasingly larger roles. For example:

• AI copilots in surveillance drones to assist human operators 

• Networked drone swarms that can saturate enemy defenses 

• Cooperative formations of undersea drones guarding critical infrastructure, all communicating and reacting faster than humans can.

Battlefield Decision-Making and Situational Awareness

In modern war, information superiority is a deciding factor. AI is being leveraged to fuse data from various sensors and sources into a coherent picture and to help commanders make faster, better decisions. This is particularly evident in Ukraine, where the large volume of battlefield data has exceeded human processing capacity. 

Drones, satellites, radars, cameras, signal intercepts, and reports from troops – all generate massive streams of raw data. Ukraine’s military addresses this by using AI-driven battlefield management systems to aggregate and analyze incoming data in near real time, providing frontline units with actionable intelligence. 

A flagship example is the “Delta” situational awareness system, which integrates inputs from dozens of platforms. As described by Ukrainian sources [9]: 

1. Delta ingests tens of terabytes of video, photos, acoustic data, open-source intelligence, and text data from sources across the front every day. 

2. AI algorithms then sift this data for patterns: 

   1. Computer vision models scan drone and satellite imagery to detect enemy vehicles or camouflaged positions. 

   2. Acoustic ML models analyze artillery sound signatures to locate guns. 

   3. Natural language processing (NLP) tools are applied to intercepted communications or social media posts (with tools like Griselda text analysis) to flag useful intelligence. 

3. The results are: 

   1. Precise locations of enemy units. 

   2. Identified equipment types. 

   3. Movement updates. 

These insights are pushed directly to field units through tablets and phones, ensuring rapid access to up-to-date battlefield information. This gives nearly every Ukrainian commander and many soldiers a common operating picture (COP) of the battlefield, updated in real-time. Such AI-enhanced situational awareness is crucial for coordinating defense and directing strikes. For instance, when a drone spots a Russian tank, an AI system can recognize it and geolocate it within seconds, immediately alerting nearby infantry or automatically cueing an artillery fire-control system. This dramatically shortens the sensor-to-shooter timeline.

Current challenges remain, of course. Ukraine’s analysts note that humans alone struggle with the “three Vs” of modern ISR data – volume, velocity, and variety. Manually reviewing endless hours of drone video or combing disparate intel reports leads to fatigue and missed cues. There are inherent delays when intelligence has to pass through human analysts before reaching shooters, and it’s difficult for humans to synthesize multi-source data, such as correlating a thermal drone image with an electronic intercept and a report from a spy on the ground quickly enough in the heat of battle. AI is helping to overcome these limitations.

Effective Training Datasets for Defense Applications

AI models vary depending on the task – some prioritize speed for real-time alerts, others focus on precision in dense or cluttered scenes. Whether scanning for vehicles, artillery, or trench lines, these models are now embedded directly into drones and edge devices. Performance hinges on training with diverse, accurately labeled, and relevant datasets that mirror operational conditions, such as varying terrains, weather, and sensor types. Synthetic data generation is increasingly used to supplement real-world data, addressing the challenge of obtaining sufficient labeled examples.

Similarly, signal intelligence is being turbocharged by AI: patterns in radio communications or radar emissions that would take an expert hours to discern can be spotted in seconds by machine learning. Cyber specialists use AI to filter intercepted enemy messages from group chats, extracting keywords about troop movements, which, when fused with location data, reveal an impending attack.

On the NATO side, these trends align with the concept of Joint All-Domain Command and Control (JADC2), which is a vision of seamlessly linking sensors, shooters, and decision-makers across all domains (air, land, sea, cyber, space). AI is seen as a backbone for JADC2, needed to automate the data processing and help connect the dots between myriad Allied systems.

The U.S. Department of Defense’s Project Maven is illustrative. Launched in 2017, Project Maven applies AI and machine learning to the flood of aerial surveillance footage collected by drones [4]. It uses algorithms to detect and classify objects in live video, alerting analysts to potential targets. Maven significantly reduced the manpower needed to monitor drone feeds and sped up the “kill chain.” During the Russia-Ukraine war, Maven helped process satellite imagery for Ukrainian forces but faced challenges with complex conditions like snow or foliage, requiring human verification to ensure accuracy.

These lessons are informing NATO’s approach: AI can greatly accelerate routine or well-defined ISR tasks, but human expertise remains vital for judgment, especially in messy real-world environments. The goal is human-machine teaming, where AI crunches data and offers insights or recommendations, while human commanders make the final decisions, guided by AI but not dictated by it [1].

The Alliance’s Defence Innovation Accelerator (DIANA) and its €1 billion Fund are targeting solutions that, for example, use AI to automatically connect Allied intelligence feeds or to create coalition-friendly data formats for battlefield information [10].

Logistics and Infrastructure: Predictive Maintenance and Autonomous Resupply

Military logistics – the coordination of troops, fuel, ammunition, and supplies – is increasingly shaped by AI tools. Two areas in particular are seeing major changes:

• Predictive maintenance: using AI to keep equipment running and reduce downtime. 

• Autonomous resupply: using AI-assisted unmanned vehicles to deliver supplies in dangerous or hard-to-reach areas.

Predictive Maintenance: Armed forces operate vast fleets of vehicles, aircraft, ships, and other equipment that demand constant upkeep. Traditionally, maintenance is either done at fixed intervals or reactively when something breaks. AI enables a shift to condition-based maintenance, where algorithms analyze sensor data and usage patterns to predict failures before they happen. Modern military platforms are heavily instrumented – tracking engine temperatures, vibrations, fuel usage, and more. By applying ML to historical data (e.g., a tank’s vibration signature correlated with breakdowns), AI systems can flag subtle signs of wear before failure occurs.

NATO notes that predictive analytics and digital twin technologies now let militaries anticipate failures and schedule repairs proactively [1]. For instance, an AI might warn that a helicopter gearbox has an 80% chance of failing within 10 flight hours – allowing crews to swap it during routine servicing rather than risking in-flight failure. U.S. Air Force programs have already shown success here, and NATO allies are adopting similar systems.

By 2025, many allied militaries are piloting AI platforms that aggregate maintenance logs and sensor feeds, issuing alerts like:

“Tank #203 track tension abnormal. Possible suspension issue. Inspect within 20 hours.”

This not only prevents equipment loss in combat but also improves sustainment planning – predicting which parts will be needed and where.

Autonomous Resupply: Delivering ammunition, food, medicine, and fuel to the front lines is dangerous – supply convoys are frequent targets. AI-powered autonomous vehicles offer a way to reduce risk and keep soldiers out of harm’s way. NATO armies have pursued this aggressively, especially after years of convoy ambushes in Iraq and Afghanistan. The concept includes self-driving trucks that follow a lead vehicle (platooning), unmanned ground vehicles, cargo drones, and autonomous helicopters dropping supplies to forward units.

Ukraine’s logistics still depends heavily on human drivers, but local programmers are developing autonomy behind the scenes. In one test, coders repurposed self-driving car algorithms to let a convoy of unarmored supply vehicles follow a military GPS route autonomously. Even basic AI tools that prioritize supply deliveries based on battlefield need have helped Ukraine optimize scarce resource allocation. 

Training, Simulation, and Cognitive Readiness

• AI-Enhanced Training: AI is transforming how soldiers train and maintain cognitive readiness. Integrated with Virtual and Augmented Reality (VR/AR), these systems create immersive, adaptive environments that tailor exercises to individual needs [13]. NATO forces now rely on high-fidelity simulators – from pilot trainers to VR shoot houses – where AI-controlled adversaries react to trainees’ tactics, forcing real-time adaptation.

• Extended Reality (XR): In Europe, XR platforms are pushing realism even further. In late 2024, Germany’s Rheinmetall and Hologate unveiled a simulator that combines VR, AR, and physical effects to replicate battlefield conditions. A VR driving trainer, for example, simulates terrain, weather, and IED threats, with AI managing enemy actions on the route. These stress-based scenarios aim to build decision-making under pressure – before soldiers face it for real.

  XR systems now support a wide range of tasks: flight training, small-unit tactics, fire coordination, and mission planning. AI also monitors performance – tracking eye movement, measuring reaction time, and highlighting blind spots or hesitation, helping instructors refine drills.

• AI-Assisted Planning and Wargaming: Traditionally, commanders trained through tabletop exercises or computer-assisted war games staffed with human role-players. Now, AI-driven simulations can fill those roles. NATO’s Modelling and Simulation Group has showcased systems where AI simulates civilian populations reacting to military actions – fleeing, hiding, posting on social media – forcing officers to consider collateral effects. AI can also generate realistic scenarios. Given an area of operations, it might predict likely enemy actions and simulate outcomes, helping commanders test and refine plans. These tools train strategic thinking and expose weak points that might otherwise be missed.

• In Ukraine, the urgency of war has accelerated training timelines. AI-based automation has simplified instruction for complex systems – especially drones. Early in the conflict, drone training took weeks. Now, with AI autopilots and targeting aids, even novice operators can be effective in hours. Some programs report that autonomous drone training can be completed in just 30 minutes, vastly expanding the pool of deployable personnel [3].

• Meanwhile, AI tutoring systems are emerging for tasks like language training and evaluating psychological resilience. NATO research projects are also exploring AI tools that monitor cognitive load in real time using biosensors — tracking heart rate, brainwaves, and more during intense training. The goal is to personalize stress exposure: pushing trainees far enough to improve decision-making under pressure without tipping into overload. Though still early, these systems could evolve into AI programs that calibrate challenge levels with precision. The following table outlines AI technologies used in VR/AR simulations for defense applications.

Cybersecurity and Cyber Defense

Cybersecurity in the defense context includes protecting military networks from intrusion, detecting and countering cyberattacks (including AI-augmented ones), and safeguarding the information domain – such as combating AI-driven disinformation.

On the defensive side, AI tools are used to monitor networks, detect anomalies, and respond to threats far faster than human analysts. Traditional signature-based defenses often fail against evolving threats. ML methods, like anomaly detection, learn what “normal” traffic looks like and flag deviations that suggest a breach or malware.

NATO’s cybersecurity strategy emphasizes that the scale and speed of modern cyberattacks demand automation. A military network might generate millions of log events daily – AI systems can triage them, flagging patterns like unusual login attempts or odd-hour data exfiltration in seconds. NATO also maintains a Malware Information Sharing Platform, where AI helps analyze and share malware signatures across Allies, ensuring one nation’s discovery benefits the rest.

In 2024, NATO leaders approved a new Integrated Cyber Defense Centre (launching by 2028) to unify threat intelligence and AI-driven cyber defense across the Alliance. AI is being used for:

• Attribution – identifying attackers by analyzing digital fingerprints and tactics

• Cyber deterrence – spotting intrusions quickly and kicking adversaries out before damage is done

Within Ukraine’s security service, AI is used to analyze large amounts of captured enemy malware code to understand its behavior quickly. As noted earlier, Ukraine also integrates cyber intelligence with its kinetic operations. For example, an AI might sift through a leaked Russian database to find personnel records that identify occupation collaborators, aiding counterinsurgency efforts. Ukrainian cyber officials have hinted that they use AI to hunt threats inside their networks. Still, details are classified [12]. What is publicly known is that Ukraine has a dedicated government cybersecurity department, the State Service of Special Communications and Information Protection (SSCIP). It has also warned that Russia is increasingly using AI in its operations to:

• Automate recon on targets 

• Generate more realistic fake messages, including deepfake audio or video messages powered by GenAI methods

• Speed up vulnerability discovery 

Generative AI in Defense Cybersecurity

Generative AI is emerging as a dual-use technology, reshaping defense cybersecurity from threat detection to training. The table below outlines key GenAI applications in this domain.

On the technical front, defensive AI is being trained on attack data to recognize the fingerprints of AI-powered hacks. For example, an AI system might detect when phishing emails are too perfectly crafted, a sign of machine-generated text, by noting statistical anomalies. Human-written messages often have certain rough edges that AI-generated text may lack. 

Cyber defenses also increasingly run automated wargames: AI-assisted “red teams” simulate attacks on friendly networks to probe for weaknesses, allowing defenders to patch vulnerabilities proactively. Ukraine’s cyber teams, working with partners, likely run continuous drills, where AI tools attempt to breach their own critical systems, such as energy grids, so they can fix any gaps before Russia finds them. 

Another facet is securing AI systems themselves from cyber threats. As armies put AI into drones, missiles, and decision aids, they must prevent adversaries from hacking or spoofing those AI algorithms. NATO’s baseline requirements for AI in defense include robustness against manipulation, such as adversarial examples that could trick an AI. Researchers are developing AI that can detect if it’s being fed malicious data and refuse or flag it, crucial so that, say, an enemy cannot fool a target recognition AI with a cleverly painted tank. Ukraine has taken steps, such as encrypting the AI software on drones, to make it difficult for adversaries to copy if a drone is captured [3]. 

In the information space, AI is also being used to counter propaganda and deepfakes. NATO StratCom (Strategic Communications) units employ AI tools to scan social media for false narratives and attribute bot activity. For example, during the Ukraine war, if a fake video emerges, AI-based media forensics can often quickly analyze it and determine if it’s likely fake by detecting inconsistencies in pixels or audio that are imperceptible to humans. This capability is vital as enemies use AI to generate disinformation at scale. NATO’s updated strategy explicitly mentions protecting our societies from AI-generated disinformation as a priority [5]. So, cyber defense in a broad sense now includes what some call “cognitive security”, meaning defending the minds of soldiers and the public against AI-augmented influence operations. [12]. 

Conclusion

From autonomous drone swarms to predictive maintenance, AI is reshaping the hardest problems in defense. Ukraine has become a proving ground – showing how drone autonomy can stall a larger force, how connected intel systems can guide pinpoint strikes, and how a digitally fluent military can punch above its weight.

NATO’s updated AI strategy focuses on interoperability, responsible deployment, and resilience against hostile AI. The alignment with Ukraine is clear: both aim to save lives by keeping soldiers out of harm’s way, act faster than the enemy, and stay adaptive under pressure.

AI has begun to fulfill its promise as a game-changer in defense, but it hasn’t replaced the human element. It’s augmenting it – whether it’s a Ukrainian drone operator achieving results that once required an entire airstrike team, or a NATO cyber analyst defending networks with an AI co-pilot.

From the editor: The problem is, in wartime, we don’t know what’s happening behind the closed doors of uncooperative states. While democracies weigh ethics and oversight, authoritarian regimes may move with fewer constraints — racing to weaponize AI in ways the rest of the world won’t see until it’s already in play. Now, everything moves at the speed of code.

Sources and further reading