NATO Startup Scaleout Gives Small Drones Autonomous Targeting AI
Scaleout Systems puts compact AI on drones for autonomous target selection, federated battlefield learning and missions despite communications jamming.
Summary
Scaleout Systems, founded by Uppsala University researchers in Sweden in 2018, shifted from deploying machine learning on commercial trucks and vehicles to defense after Russia’s 2022 full scale invasion of Ukraine. CEO and cofounder Andreas Hellander says its compact computer vision models replace reliance on frontier systems from OpenAI or Anthropic, running on hardware ranging from embedded drone devices to powerful edge workstations. NATO selected Scaleout for its Defence Innovator Accelerator for the North Atlantic, DIANA, Challenge Program in 2025, where its Federated Aerial Intelligence for Recon project equips drones, pilot tablets and field command posts to identify targets, send selective updates without sensitive raw data, retrain models at platoon or company headquarters, and distribute improvements during a day or operation. Local processing preserves capability during jamming, communications loss or attacks on centralized data centers, including facilities destroyed during the US and Iran war.
Scaleout publicly demonstrated a surveillance and reconnaissance system in Sweden and joined BAE Systems Bofors’ Affordable Loitering Modular Ammunition, ALMA, project for a low cost autonomous kamikaze drone. At Winter Demo 2026 in January, onboard AI detected, identified and geolocated threats in real time, prioritized an armored engineering vehicle, and independently flew over it to drop an explosive, although a human operator retained control. In June 2026, a Swedish Air Force base in Uppsala, where the military licenses Scaleout’s main platform, continued inference and active learning after disconnecting from the company’s laboratory, then synchronized updates after reconnection. Scaleout says this federated approach could expand across regions, countries and NATO members.
Positives
- NATO selected Scaleout Systems for its DIANA Challenge Program in 2025 to advance distributed aerial intelligence.
- Winter Demo 2026 showed ALMA detecting, identifying and geolocating threats in real time using only onboard computing.
- June 2026 testing preserved AI inference and active learning after a Swedish Air Force base lost its laboratory connection.
- Selective model updates let headquarters improve battlefield AI without receiving sensitive raw sensor data.
- Federated learning could eventually support model collaboration across regions, countries and NATO members.
Risks & concerns
- ALMA independently prioritized an armored engineering vehicle and dropped an explosive without direct human commands.
- Models trained in desert conditions may perform poorly when deployed in urban environments.
- Enemy jamming and electronic warfare can disrupt the communications needed to exchange model updates.
- Centralized AI infrastructure remains vulnerable, as data centers were targeted and destroyed during the US and Iran war.
