Desafíos actuales de la Inteligencia Artificial

C apítulo 5 ANALYSING THE INTERPLAY BETWEEN DATA SPACES AND ARTICLE 10 OF THE AI ACT: A CASE STUDY OF CREDITWORTHINESS AI SYSTEMS Andrés C homczyk P enedo Affiliated researcher. Law, Science, Technology and Society research group. Vrije Universiteit Brussel. Anna C apellà i R icart Postdoctoral researcher. Institute of Law and Technology (IDT). Universitat Autònoma de Barcelona. ABSTRACT: This paper explores the challenges Article 10(5) of the AI Act faces in ensuring that special categories of personal data can be used to mitigate biases in AI datasets. Article 10(5) imposes strict limitations on processing such data, particularly regarding data sharing. Simultaneously, the EU is promoting data spaces, i.e., common ecosystems for data sharing to facilitate data-driven innovation. This presents a potential clash between regu- latory objectives. As such, this paper provides a comprehensive understanding of the legal challenges in balancing the need to access personal data to develop unbiased AI systems while limiting access to such data in a context where data is expected to flow freely. Focusing on financial services, the paper examines creditworthiness AI systems as a case study. These systems, labelled high-risk and therefore subject to Article 10(5), can benefit from specific data sharing rules within the financial data space, also known as open finance. Through this case study, the paper illustrates the complex interplay between regulatory requirements and the practicalities of data sharing in high-risk AI systems, offering insights and recommendations for policymakers and stakeholders. KEYWORDS: data spaces, algorithmic bias, discrimination, financial services, credit- worthiness