AI system — AI-rendszer

The Hungarian text of the Regulation refers to this as an „MI-rendszer”. The definition in Article 3(1) has six elements: machine-based operation, varying levels of autonomy, the capacity to adapt after deployment (an optional element — its absence does not rule out the classification), explicit or implicit objectives, inference from input, and outputs that influence physical or virtual environments. What separates this from conventional software is inferential capability: recital (12) makes clear that if the rules are defined solely by natural persons and the system only executes them automatically, it falls outside the concept — a payroll program or a rule-based alert is not, by itself, an AI system. The Commission's guidelines of 6 February 2025 (C(2025) 924 final) place four software types — mathematical optimisation, basic data processing, classic heuristics and simple prediction — typically outside the concept, though not categorically; the guidelines are non-binding interpretive guidance, not law. Crucially, the classification only opens the gate: being an AI system does not by itself trigger any substantive obligation — that depends on the risk category (prohibited practice, high-risk, subject to transparency duties, or none of these).