AI system (Article 3 EU AI Act)
Machine-based system with a degree of autonomy under Article 3(1) AI Regulation
TL;DR
Under Article 3(1) of the EU AI Act, an AI system is a machine-based system that operates with varying levels of autonomy, may exhibit adaptiveness after deployment, and produces inferences from inputs (predictions, content, recommendations, decisions).
What is an AI system (Article 3 EU AI Act)?
The definition was clarified by Commission Guidelines C(2025) 5053 final of 29 July 2025. Seven core elements of the definition: (1) machine-based system, (2) degree of autonomy, (3) adaptiveness after deployment (possible, not mandatory), (4) inputs, (5) inferences, (6) explicit or implicit objectives, (7) influence on physical or virtual environments. Key distinctions: Classical rule-based software is NOT an AI system. Standard statistical analysis tools (e.g., Excel pivot tables) are also excluded. A simple forecasting model using linear regression: borderline case — the guidelines consider the 'learning character' to be decisive.
Practical example
Clear examples of AI systems under Article 3: - Generative AI (ChatGPT, Claude, Gemini) - Image recognition (e.g., defect detection in production) - Recommender systems (Netflix, Amazon) - HR recruiting algorithms - Predictive maintenance models Clear non-AI: - Excel formulas and pivot tables - Classical SQL reports - Rule-based workflows without a learning component