Vulnerability in the EU AI Act: building an interpretation

ACM Conference on Fairness, Accountability, and Transparency (FAccT), pp. 1985–1997

Abstract

The EU Artificial Intelligence Act (AIA) integrates the concept of (human) vulnerability as a critical element for categorising high-risk AI systems, prohibiting exploitative practices, and enforcing safeguards for fundamental rights. Yet the Act lacks a unified definition, leading to fragmented and inconsistent application. This paper addresses this gap by proposing a comprehensive framework for understanding vulnerability in the AIA, rooted in interdisciplinary scholarship and EU legal traditions. We argue that vulnerability should be understood as a heightened risk to fundamental rights arising from power imbalances, dependencies, and varying levels of personal or structural resilience. Through a systematic analysis of the AIA's vulnerability-related provisions, we demonstrate how this framework can harmonise its diverse references, inform risk assessments, and guide regulatory enforcement.

Authors 2

1. Maria-Lucia Rebrean Aachen eLaw Center for Law and Digital Technologies

   Leiden University

   Affiliation as printed

   Leiden University, eLaw Center for Law and Digital Technologies, Leiden, Netherlands

2. Gianclaudio Malgieri Aachen eLaw Center for Law and Digital Technologies

   Leiden University

   Affiliation as printed

   Leiden University, eLaw Center for Law and Digital Technologies, Leiden, Netherlands