A Default Model of Normative Risk for the EU AI Act

The EU Artificial Intelligence Act (AIA) aims to provide a uniform legal framework that facilitates the development, marketing, and use of AI systems in the EU. However, in its current state, important adjustments are necessary for it to fully achieve its intended goals. As Novelli et al. (2024) have already noticed, the AIA needs to be complemented with a methodology that can help the relevant actors (in particular, AI providers and deployers) to determine the appropriate risk category of a particular AI system to identify the proportional obligations associated to it. Yet, while their proposal to address this gap is interesting, it is argued here that their view fails to offer a feasible model. The main issues stem from their commitment to both an overly particularist understanding of risk assessment (i.e., that the correct risk model must consider the interplay of multiple factors in specific situations) and a contested method of constitutional adjudication (i.e., that a proportionality test is indispensable for balancing competing fundamental rights). With the intention to submit a more compelling alternative, the paper builds on a default theory of practical reasoning to advance a general framework for decision-making in risk scenarios. The purpose of supplying the AIA with this methodology is to introduce a fine-grained approach to the application of its harmonised directives, especially in relation to the requirement of conducting adequate risk assessments and implementing the corresponding risk management measures, without compromising its generality and flexibility.