AI & the future of defence

A modern unit receives information from thousands of sources at once: radar, drone cameras, satellite images, intercepted signals and reports from its own troops. AI fuses those streams into one current picture, recognises vehicles and movements, and lists possible targets with the reasoning behind them. Where an analyst once searched for hours, a staff now sees much sooner what is changing. NATO also works with such an AI-supported system for its operational picture. The next step is a picture of what is coming: AI that uses patterns in movements, logistics and signals to estimate how likely the opponent's next move is, with the uncertainty attached. People set the mission and the limits; within those limits, some weapons select their own targets.

Open sources produce more than any team of analysts can take in: social media, shipping and flight data, commercial satellite imagery. AI searches that stream in many languages and links a video of a convoy to a place, a time and a satellite image of the same road. AI agents now also look for additional sources on their own; according to one supplier, NATO is among the first users. In a scenario for 2040, cooperating agents draw different explanations for a military movement from such data, test them continuously against new observations and work through possible next steps. The analyst then sees how the expected course of events shifts as soon as new information contradicts an earlier explanation, and remains the one who interprets it.

Small drones in large numbers have quickly become a fixture of armed forces: they scout, correct artillery fire, deliver supplies and are also used as weapons. That drives demand for counter-drones. AI plays a part on both sides. In the drone, it helps recognise targets and hold a course when the link drops. In defence, it combines radar, cameras and radio signals, so a small drone over a barracks, port or airfield is quickly told apart from a bird. Interceptor drones, which bring down a hostile drone themselves, are already in use. The race is shifting towards deploying them at larger scale, with more autonomy in searching, tracking and intercepting.

A staff prepares plans and decisions, often under heavy time pressure. In trials, AI takes on a growing share of the analytical work. It helps develop strategies and possible plans, tests them against different moves by the opponent and compares them on risk, time and resources. During execution, it monitors whether the assumptions behind the plan still hold and reports where reality diverges. Armed forces test these systems on closed networks, so classified information stays inside. A staff can examine more options in the same amount of time. That leaves more time for the work people are good at: making choices, weighing what the numbers leave out and working with people.

Military networks connect command systems, radar, aircraft and ships full of software, and every connection is a possible way in. AI monitors large volumes of traffic and spots anomalies a person would miss, such as a system that suddenly starts sending data in the middle of the night. It is also moving to the source: in a competition run by a government, AI systems showed they could find and repair weaknesses in software on their own. The next step is cyber defence in which AI agents on both sides attack and patch at machine pace. People then set the priorities: which system goes offline, and which has to keep running during an operation. Whoever cannot keep up with that pace has less and less time to defend.

Seabed cables carry a large share of international data traffic and also transmit electricity; pipelines carry gas and oil. They run for thousands of kilometres and are largely unguarded. AI links the behaviour of ships to the position of those connections: it combines the signals ships transmit with radar, satellite images and data from underwater drones, and raises the alarm when a ship moves unusually slowly, lingers above a cable or switches off its beacon. After several incidents, NATO has stepped up its watch over the cables in the Baltic Sea, with ships, aircraft and uncrewed vessels. The next step is a seabed that guards itself: underwater drones on permanent patrol and fibre-optic cables that report vibrations, acting as sensors themselves.

A unit is only as deployable as its equipment and its supply chain. With predictive maintenance, sensors in vehicles, aircraft and ships measure the condition of parts, and AI works out when something is about to fail. The mechanic replaces the part before it breaks down. The same principle applies to supply: AI helps estimate how much ammunition, fuel and spare parts an operation needs and where they should be. Then there is the 3D printer, which makes a part in a container close to the operation that would otherwise take weeks to arrive. Part of supply shifts from transporting things to making them on the spot.

In simulators, military personnel practise what would be too costly, too dangerous or too rare to rehearse for real, from a dogfight to an operation across an entire neighbourhood. AI makes that training more realistic: the opponent follows no fixed script but reacts to what the trainees do. In a simulation, an AI pilot beat an experienced fighter pilot five times in a row; later, AI also flew a real test aircraft in a practice dogfight. The next step is training in a digital twin of the real area of operations, against an AI opponent that already knows the tactics of recent weeks. Lightweight AR glasses then merge field exercises with virtual ones: the soldier moves through real terrain where virtual opponents appear.

In a hybrid threat, a state combines different means: influence, cyber attacks, economic pressure, sabotage and sometimes military action. It often stays below the threshold of war, using means that each look small on their own: a suspicious fire, drones over an airfield, a wave of fake news. Each incident lands with a different agency. AI helps lay those scattered signals side by side and see patterns: thousands of accounts spreading the same story at once, or a series of incidents pointing to one plan. A pattern does not yet prove who is behind it. Generative AI makes influence personal and possible at scale; at the same time, AI can sharply shorten the time between the first signals and recognising a coordinated campaign.

In AI-first armed forces, sensors, AI and autonomous systems do much of the repetitive and predictive work that people do not need to do. Military personnel concentrate on judgement, contact and responsibility. That affects how teams are made up and how a staff works. For an organisation that struggles to recruit, it is also a practical choice. A system can watch an area for days without tiring. Weighing consequences, building trust and carrying responsibility stay with people. In a scenario for 2040, new teams are designed from the drawing board around what machines do, and one person directs several systems.

As AI and uncrewed systems take over much of the hands-on work, military personnel shift from operators to directors. Military think tanks sketch the jobs that come with it. The swarm leader does not fly one aircraft but gives hundreds of systems a goal and limits. The model trainer in the field feeds AI with new data from the area of operations, so a recognition model learns to identify a new enemy vehicle within days. The spectrum strategist matches communications, sensors and jamming capacity to the manoeuvre. The cognitive defence specialist protects units and the population against influence operations. Reservists bring knowledge from the tech sector. What these roles share: the edge depends more and more on brainpower and adaptability, and less on numbers.

In a mixed unit, soldiers work alongside uncrewed systems and AI. An infantry section might get drones that scout ahead, a robotic vehicle that carries loads or evacuates the wounded, and AI that sorts incoming reports. What the machine does well, the machine does, such as watching an area for hours. The soldier does what calls for judgement: choosing, talking to the local population, stepping in. Exercises show how much predictable behaviour and understandable reports shape the cooperation. The next step is a unit with several drones and robots per soldier. The section leader then also leads a team of machines, deciding what each system may do on its own and when a person takes over.

A future academy is an in-house training centre that prepares staff and leaders for tomorrow's work. New AI applications can change the work before training programmes have caught up, and the organisation does not wait for them. Current staff learn to work with AI and new technology there, and new colleagues practise with the organisation's own applications from day one. The academy draws its course material from the organisation's own work, and changes as that work changes.

No organisation develops all its technology alone. Innovation emerges in an international ecosystem of partners, start-ups, scientists and friendly governments. How well that ecosystem works determines how quickly an innovation reaches the people who use it. Start-ups build software and hardware at a pace that traditional procurement cannot match; organisations that give them room to test, data and small contracts gain access to that pace. Governments pool research funding, test each other's solutions and buy together. An invention from one country can then be in use in another within months.