NATO Drone Demonstrates Autonomous Target Selection

During a demonstration in January, a loitering munition developed by a Swedish company identified an armored vehicle on a snowy test range and engaged it without a human operator selecting the specific target. The system, created by Scaleout Systems, a spin-off from Uppsala University, detected multiple objects, ranked the armored engineering vehicle as the highest priority, and executed the strike autonomously. This event illustrates a growing trend in defense technology where artificial intelligence is moving from assisting human operators to managing the final stages of engagement independently.

The technology was showcased under the ALMA project, led by BAE Systems Bofors, which focuses on affordable and modular unmanned systems. According to reports from Ars Technica, the drone used dedicated onboard computing to handle detection, identification, and geolocation, ensuring that no data was sent to external servers. This architecture is designed to withstand electronic warfare and communication jamming, allowing the aircraft to remain operational even when cut off from ground control. The demonstration was part of a larger event in Karlskoga, Sweden, involving numerous technology firms and attended by Swedish defense officials.

Onboard Intelligence Enhances Battlefield Resilience

Scaleout Systems, which was selected for NATO’s Defence Innovation Accelerator in 2025, emphasizes the importance of decentralized processing. In a separate test at a Swedish Air Force base in Uppsala, the company demonstrated that its software could continue learning and processing data after losing contact with its central lab. This capability is crucial in modern conflicts where communication links are frequently disrupted. By keeping intelligence on the airframe, the system reduces latency and prevents the loss of critical data if the connection is severed.

The approach mirrors trends seen in other recent conflicts, where unmanned aerial vehicles increasingly rely on local processing to maintain functionality. According to GN geopolitics/nato, this shift is part of a broader European effort to develop robust unmanned military systems. The company’s CEO, Andreas Hellander, noted that this technology could facilitate collaboration among NATO member states by providing a standardized framework for autonomous operations. The system uses smaller, specialized computer-vision models rather than large frontier AI, optimizing performance for the limited hardware available on drones.

Federated Learning Addresses Model Drift

One of the primary challenges in deploying AI on the battlefield is model drift, where a system trained in one environment fails to perform effectively in another. Scaleout’s solution involves a federated learning architecture, where devices share model updates rather than raw sensor data. This method allows headquarters to retrain models on aggregated battlefield data and push updates to edge devices when communication links are restored. By avoiding the transfer of raw footage, the system also helps protect reconnaissance data and ensures that sensitive information remains on local hardware, addressing both performance and security concerns.

Autonomy Raises Regulatory and Ethical Questions

As autonomous systems become more capable, the international community faces complex questions regarding the regulation of lethal autonomous weapons. While the technology offers tactical advantages in jammed environments, it also challenges existing norms of human oversight in warfare. The demonstration in Sweden highlights the rapid pace of development in this field, with companies like Scaleout moving quickly to integrate AI into standard defense platforms. Observers will likely watch how NATO member states respond to these advancements, both in terms of adoption and the potential for new diplomatic discussions on the limits of autonomous engagement.