Small AI models let drones autonomously identify and attack ...

Scaleout Systems, a Swedish startup originally founded in 2018, is adapting decentralized machine learning for military applications following the 2022 Russian invasion of Ukraine. Selected for NATO's DIANA Challenge Program in 2025, the company utilizes small, lean AI models that run on edge hardware like drones, pilot tablets, and field command posts. This technology allows military devices to perform tasks such as computer vision and target detection directly on site without needing frontier AI models from companies like OpenAI or Anthropic.

The system relies on federated learning to continuously improve AI capabilities using data gathered during operations. Drones and field devices process sensor data locally and share selective updates with local platoon or company headquarters to retrain the models for new environments, such as shifting from desert to urban settings. This decentralized approach avoids the vulnerabilities of relying on central data centers, which face risks from military targeting and electronic warfare jamming that disrupts communication signals.

Scaleout is participating in the Affordable Loitering Modular Ammunition project led by BAE Systems Bofors, creating low-cost autonomous attack drones. During public demonstrations in Sweden, a drone used onboard computing to autonomously detect, identify, and geolocate threats in real-time, prioritizing a high-value armored engineering vehicle for an airburst attack. A human operator remains in the loop to verify targets and authorize the drone actions. Additional tests at a Swedish Air Force base demonstrated that local nodes can run inference and active learning independently even when disconnected from central laboratory servers, later syncing updates once connections are restored.