In February 2025, the European Commission adopted a set of guidelines aimed at supporting stakeholders—including policymakers, companies, and developers—in defining whether a system qualifies as AI under the legal framework of the Regulation (EU) 2024/1689, which came into force in August 2024.
EU - European Commission - Guidelines on AI system definition to facilitate the first AI Act’s rules application
Anno 2025
The guidelines emphasize the need for a definition of artificial intelligence that is both precise and flexible to ensure compliance with the regulatory framework, while balancing rapid technological development with the protection of fundamental rights, health, safety, democracy, and the rule of law. The present guidelines are relevant only for those systems fulfilling the definition given by art. 3(1) of the Regulation (EU) 2024/1689 and they do not provide an exhaustive list of all potential AI systems covered by the EU Regulation.
The seven key elements that define AI systems include:
- machine-based: the AI system must be computationally driven and based on machine operations. The machine-based nature involves both hardware and software components to enable AI systems functioning and covers quantum computing systems;
- autonomy: the AI system must have the inference capacity, which is a key element to bring about its autonomy, and it is capable to operate with limited or no human intervention in specific use contexts, demonstrating some reasonable degree of independence of actions and interacting with its external environment. This means that systems which are design to operate solely with full manual human involvement and interventions are outside the scope of AI definition;
- adaptiveness: the AI system may have self-learning capabilities and make its behaviour change while in use. The adaptiveness capability of AI tools is not a mandatory characteristic for a system to be considered artificial intelligence;
- objectives: the AI system is designed to operate according to defined objectives, which may be explicitly or implicitly defined. Explicitly objectives are clearly stated goals that are directly encoded by the developer into the systems. Implicit objectives refer to goals that are not explicitly stated but may be deducted from the behaviour or underlying assumptions of the system. The objectives of an AI systems are essential in guiding AI behaviour and interactions and they may be different from the system’s intended purpose of use;
- capacity of inference: the AI system must be able to infer, from the output it receives, how to generate outputs. This is one of the essential features to define when a system is AI and it should be understood as referring to the building phase, wherein a system derives output through AI techniques enabling inferencing. The techniques that enable inference, making AI differ from traditional software, are data processing techniques, such as machine learning (supervised, unsupervised, and reinforcement learning) and logic-and knowledge based reasoning;
- outputs that can influence physical or virtual environments: the ability of the AI system to generate outputs, such as predictions, content, and recommendations. Predictions are an estimate about an unknown value (output) from known values supplied to the system (input) that require the least human intervention. Contents refer to the generation of new materials by an AI system. Recommendations are suggestion to specific actions, products or services to users based on their preferences, behaviours or their data input. Decisions refer to conclusions or choices made by the AI system;
- interaction with the environment: the AI system actively impact the environments in which they are deployed. The influence of an AI system may be both to physical object and to virtual environments;
The seven key elements set out in that definition are not required to be present continuously throughout all phases of the AI system lifecycle.
The guidelines also clarify which systems are outside the scope of AI Act definition, such as systems for improving mathematical optimization, basic data processing, systems based on classical heuristics and simple predictions systems.
The guidelines clarify their non-binding nature, since any authoritative interpretation of the AI Act may ultimately only be given by the Court of Justice of the European Union.
The guidelines are available at the following link and in the download box.
Marta Fasan
Pubblicato il: Giovedì, 06 Febbraio 2025 - Ultima modifica: Mercoledì, 05 Novembre 2025
