Self-Service or Not? How to Guide Practitioners in Classifying AI Systems Under the EU AI Act

A new research paper titled 'Self-Service or Not? How to Guide Practitioners in Classifying AI Systems Under the EU AI Act' addresses a critical gap in the practical application of the world's first major AI regulatory framework. Authored by Ronald Schnitzer, Maximilian Hoeving, and Sonja Zillner, the study investigates how industrial professionals actually navigate the EU AI Act's complex Risk Classification Scheme (RCS), which became law in August 2024. The central finding is that while the RCS provides a theoretical foundation for a risk-based approach, its real-world application is fraught with difficulty, requiring a blend of legal, technical, and domain-specific expertise that many practitioners lack.

The researchers employed a Design Science Research (DSR) methodology, conducting two evaluation phases with 78 practitioners from diverse domains. They tested a self-service, web-based decision-support tool to understand the classification process. The results highlight significant challenges in interpreting the Act's legal definitions and determining regulatory scope. Crucially, the study demonstrates that targeted support mechanisms—such as providing clear explanations of legal terms and concrete, practical examples—can dramatically improve the accuracy and efficiency of risk classification. This work provides essential, evidence-based guidance for both tool developers creating compliance software and policymakers aiming to refine the Act's implementation, ultimately making it more actionable for the businesses it regulates.