Military aims to use AI to reshape the battlefield
Technology is accelerating the processing of information from multiple data streams and easing decision-making under pressure
In a darkened room in London’s Shoreditch district, a vast screen flickers to life displaying a map of the Baltic Sea. With a few keystrokes, maritime navigation data overlays the map, showing vessels in motion. Another command brings up the routes of undersea telecommunication cables.
Almost immediately one of the vessels, a Liberian-flagged ship heading towards the BCS East-West Interlink internet cable connecting Lithuania and Sweden, appears as a red icon.
Based on the vessel’s profile data, dominAI, the AI-powered software monitoring the fictional situation, has identified the ship as a potential threat. The operator can immediately dispatch a military drone or naval vessel to intercept. The system then runs simulations of each possible response, assigning probabilities of success to guide decision-making.
The scenario is a demonstration based on a real-world events: in November 2024, the East-West interlink was one of two submarine communications cables damaged within 24 hours by actors unknown. The authorities suspected sabotage.
DominAI is the brainchild of Hadean, a London-based former gaming company now positioning its software as the command and control solution needed by the UK military for the next generation of conflict.
Britain has earmarked £1bn to build a “digital targeting web” — an effort to use artificial intelligence to fuse data from multiple sources, both civilian and military, into a single command network. This would allow commanders to make faster, better-informed decisions about how quickly to escalate a response from mere surveillance to armed intervention against rapidly emerging threats.
“AI underpins the entire military architecture, it is or will be at the heart of the targeting process,” says Sir Richard Barrons, a former British army general, who is one of the authors of the UK’s 2025 Strategic Defence Review.
Fully autonomous intelligent machines on the battlefield remain years away, but AI is already accelerating how humans process information and make decisions under pressure. The US has already deployed an AI-powered command and control system. Maven, designed by data intelligence group Palantir, was reportedly used in combat during US air strikes in Yemen in 2024. Nato has since adopted variants of the software.
The first challenge AI is trying to solve is moving data around a battlefield without cell towers, and defeating pervasive electronic jamming, says Will Blyth, co-founder and chief executive of Arondite, which has developed a smart system named Cobalt.
“A lot of innovation goes into minimising how much data needs to move.” Then AI models can process massive streams of imagery or sonar feeds, reducing dependence on tired human operators. “From there,” Blyth adds, “it’s about orchestration — reliably cueing the best available asset, whether a drone, a robot, or a human team — and doing it as quickly as possible.”
Speed, however, is only one part of the transformation promised by AI. The other is the military concept of mass — the idea that autonomous drones and robots can fill the numerical gaps in wars against larger foes.
Nato’s 4,300km eastern frontier with Russia has spurred new thinking in the alliance about how to use drones to make up for a lack of numbers. The war in Ukraine has already demonstrated a revolution in robotics, with cheap first-person-view (FPV) drones, controlled remotely by an operator, helping the country’s military counter larger Russian forces. Russian troops are similarly using drones.
The next step is autonomy — drones and ground robots that no longer need one-to-one human operators. The US Marine Corps in November posted a video of troops training with multiple quadcopter drones built by US-based Auterion, which run software that “applies real-time AI to allow soldiers to deploy swarms that operate as a single force”, according to the company.
“If you’re a four-person team covering 20km of front line in Ukraine,” says Dean Jones, co-founder of Gallos, “you can control only so much using FPVs and robotics. [So] there’s also going to have to be an element of autonomous response.” Ukraine recently approved the semi-autonomous tracked robot “Krampus” — an armed, unmanned ground vehicle (UGV) — for battlefield deployment.
Some degree of autonomy is also critical for navigation in the face of electronic warfare jamming that can cripple direct radio control by operators. ARX Robotics, a German start up, employs UGVs in Ukraine that can navigate following waypoints with collision avoidance if communications are not possible, says David Roberts, chief executive of ARX UK.
In the skies, the UK is investing roughly £2bn per year in the Global Combat Aircraft Programme, an “optionally crewed” sixth generation fighter to be built together with Japan and Italy by 2042. It is to work together with drones, or so-called “loyal wingmen”, already being developed to fly alongside crewed aircraft.
Beneath the waves, autonomous submarines, such as Anduril’s Ghost Shark, recently sold to Australia, and submarine-detecting undersea robots, such as the SG-1 Fathom by Helsing are fast becoming a reality.
These long-endurance drones, or “gliders”, move by altering their buoyancy and have existed for decades. But Helsing has fitted its system with an AI, called Lura, to classify acoustic signatures of ships and submarines.
Yet as autonomy spreads, ethical and legal questions deepen. Can a machine truly distinguish combatants from civilians — or grasp the moral weight of so-called collateral damage?
Jessica Dorsey of Utrecht University warns that “given that humans are ultimately responsible for ensuring compliance with international humanitarian law, preserving context-appropriate human judgment is critical.” She cautions that AI systems may compress complex moral and legal judgments into “algorithmic models”.
Likewise, when machines generate hundreds of potential targets per hour, operators risk falling into “automation bias” (trusting the system by default) or “action bias” (feeling compelled to act because the system demands it). As Dorsey notes, “the challenge is not simply whether algorithmic systems can help, but how they reshape the mental architecture and process of responsible command.”
How much judgment humans are willing to surrender to the algorithm may mean the difference between victory and defeat. However, this logic may ultimately present even greater danger.