EU Commission: published the Guidelines on the interpretation of the legal concept of "artifi-cial intelligence system" pursuant to Article 3(1) of Regulation 2024/1689 (AI Act).

The interpretative action of the AI Act carried out by the EU Commission continues. After the recent approval of the Guidelines on prohibited AI practices, here is the adoption of the interpretative guidelines on the legal definition of "artificial intelligence system" (art. 3.1 Reg. 1689/2024).

These Guidelines, like those on prohibited practices, are currently approved but not yet formally adopted. These are relevant acts (although not binding) since – as mentioned in previous posts – the direct applicability as of 2 February 2025 of the first 5 articles (including Article 3, which contains 68 legal definitions) now allows not only to review contracts with providers of AI-based services and products from this perspective, but also (and this is what the latest guidelines aim at) to assist suppliers and other interested parties in the determine whether a software system constitutes an "AI system" and is therefore subject to the rules of the Regulation.

To be covered by the AI Act, we need to look at (a) the architecture of the system, (b) the specific functionalities, and (c) the presence of the seven elements mentioned in the definition in Article 3(1) of the AI Act, namely:

1. Machine-based system;
2. Autonomy;
3. Adaptability;
4. Pursuit of implicit or explicit objectives;
5. Inference and output generation using AI techniques;
6. Outputs such as forecasts, content, recommendations or decisions;
7. Outputs that may affect physical or virtual environments.