Analysing Risk-Based Approach in the Draft EU Artificial Intelligence Act

Abstract

The article delves into the risk-based approach underpinning the draft EU ArtificialIntelligence Act. Anticipated to be approved by the end of 2023, this regulation ispoised to serve as a cornerstone in the European Union’s legal framework forgoverning the development and deployment of artificial intelligence systems (AIsystems). However, the ever-evolving technological landscape continues to presentnovel challenges to legislators, necessitating ongoing solutions that will span yearsto come. Moreover, the widespread proliferation of foundation models and generalpurpose AI systems over the past year underscores the need to refine the initialrisk-based approach concept. The study comprehensively examines the inherentissues within the risk-based approach, including the delineation of AI systemcategories, their classification according to the degree of risk to human rights, andthe establishment of optimal legal requirements for each subset of these systems.The research concludes that the construction of a more adaptable normative legalframework mandates differentiation of requirements based on risk levels, as well asacross all stages of an AI system’s lifecycle and levels of autonomy. The paper alsodelves into the challenges associated with extending the risk-oriented approach toencompass foundation models and general purpose AI systems, offering distinctanalyses for each.