Desafíos actuales de la Inteligencia Artificial

Analysing the interplay between data spaces and article 10 of the AI act: a case study of ... 79 the FiDAR proposal seeks to create a more integrated and competitive financial services mar- ket by enhancing data sharing and accessibility among the different stakeholders, leading to the consolidation of an open finance framework. The FiDAR proposal intends to establish rules on the access, sharing and use of certain categories of customer data in financial services, as detailed in its Article 1. It covers a selected group of data and financial institutions that will be engaged in this data space. While dealing with financial data, the proposal does not define this concept expressly but rather deals with ‘customer data’, that is defined as personal and non-personal data that is collected, stored and otherwise processed by a financial institution as part of their normal course of business, un- der Article 3(3). Financial data could, in certain scenarios be considered as a special category of personal data (Chomczyk Penedo and Trigo Kramcsak, 2023), as it can reveal information specially protected, for example if someone donates money to a political party (revealing their political opinions) or pays a monthly contribution to a trade union. When it comes to creditworthiness AI systems and the data related to them, the FiDAR proposal adopts a particular approach. In this respect, we can distinguish between the data used to train the AI system, the data used to operate it, and the data resulting from it. In this respect, the FiDAR proposal only includes within its scope data about the creditworthiness, i.e., the data used to operate the AI system and obtain a credit score; it then furthers limits by only focusing on firms and not consumers. When it comes to the data related to consumers and their creditworthiness, Recital 18 highlights that the risk of exclusion outweighs the ben- efits from sharing data related to consumers. As for the other data categories included in the scope of the FiDAR proposal, its use for credit scoring activities shall be subject to limitations to be established by the European Banking Authority and the European Data Protection Board through the data use perimeters, as provided for under Article 7(2). While these provisions tackle the use of data by AI creditworthiness systems and the possibilities of getting access to other data categories for the conduction of a creditworthiness assessment, these do not answer how to deal with the training of the AI system itself. Moreo- ver, it also does not tackle whether certain data, that can be considered as a special category of personal data, generated by a financial data space participant can be made available in the data space for others to improve their systems. 4.3.Interplay and clashes between data spaces and detection and correction of algorithmic biases In Article 10(5) AI Act, we find no limitations regarding the origin of the data. In this respect, it could be data already collected by, based on our case study, a bank, for example; also, it could be obtained through the data space if uploaded by another participant, for example a credit bureau. Here we understand that all persons that access the data space are only authorised persons with appropriate confidentiality obligations (Article 10(5)(c)) and that data gathered from the data space afterwards cannot be accessed by other parties even