DAV Calls for Legal Clarification: Classical Statistical Methods Should Be Distinguished from AI Systems under the AI Act

Background and Position of the DAV

With Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (hereinaf-ter: AI Act), the European Union has for the first time created a horizontal legal framework for AI systems. The European Commission’s guidelines on the definition of an AI system, set out in Communication C(2025) 5053 final of 29 July 2025, are intended to give guidance for the practical interpretation of Article 3(1) of the AI Act. From an actuarial perspective, however, considerable scope for interpretation remains open, in particular for classical statistical methods such as linear and logistic regression as special cases of generalised linear models (GLMs), and GLMs as spe-cial cases of generalised additive models (GAMs). These methods have formed part of the stand-ard methodological repertoire of actuarial science for three decades and have proven their worth over the long term, for example in decision-relevant applications such as pricing or risk assess-ment. Actuaries are trained over many years in their application and thereby ensure traceability, transparency, and non-discriminatory use.

The German Association of Actuaries (Deutsche Aktuarvereinigung e. V. – DAV) takes this scope for interpretation as an occasion to set out its professional position: classical statistical methods should not be classified as AI systems within the meaning of Article 3(1) of the AI Act. This state-ment explains why this distinction follows from the legal criteria themselves and asks the Europe-an Commission to provide explicit clarification.
 

Legal Framework: The European AI Act

Decisive for the question to be assessed here is the definition in Article 3(1) of Regulation (EU) 2024/1689. It reads:
“[An] AI system [is] a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, con-tent, recommendations, or decisions that can influence physical or virtual environments.” 

Within this definition, the Commission guidelines distinguish seven main elements. For the deline-ation question at hand, four sets of characteristics in particular can be derived from the above definition: (i) the machine-based character, (ii) the design to operate with varying levels of auton-omy, (iii) possible adaptiveness after deployment, and (iv) the capability to infer, from the input received, outputs for explicit or implicit objectives. While the machine-based character is generally given owing to the analysis of large volumes of data, the characteristics of autonomy, adaptive-ness, and inference capability are particularly decisive for the delineation of classical statistical methods.

Recital 12 of the AI Act gives concrete form in particular to the characteristics of inference and adaptiveness. It clarifies that the capability of an AI system to infer “transcends basic data pro-cessing” and enables learning, reasoning, or modelling. An AI system is adaptive if it possesses a self-learning capability through which it can “change while in use”.
 

Read the full text here.