Desafíos actuales de la Inteligencia Artificial

250 Desafíos actuales de la Inteligencia Artificial 1. INTRODUCCIÓN From 2020, while the debate on the regulation of artificial intelligence (AI) by legislative bodies is still on the horizon, social security agencies in countries such as France, the United Kingdom (UK) and Brazil start to increasingly use automated and AI decision systems to as- sess the eligibility and level of social security benefits. By 2024, AI legislation is beginning to take root, and not only the regulatory legitimacy of federal agencies in relation to AI systems, but also their ongoing use, will need to be reviewed. The Artificial Intelligence Act (AIA), (Regulation (EU) 2024/1689, approved by the Eu- ropean Union (EU), prohibits “general purpose” scoring systems that could reduce individ- uals to a single social score that affects various aspects of their lives (Article 5, c, AIA). While the AIA classifies as “high risk” (Annex III(5)(a)) “AI systems intended to be used by public authorities or on behalf of public authorities to evaluate the eligibility of natural persons for essential public assistance benefits and services, including healthcare services, as well as to grant, reduce, revoke, or reclaim such benefits and services” it does not fully cover current AI scoring practices that assess social security benefits. Therefore, while it’s clear that systems that evaluate and score natural persons’ data to decide on social security benefits are classified as high risk, as they pose a significant risk (Article 6, 3, AIA) to the fundamental right to social security recognised by the Charter of Fundamental Rights of the European Union (Article 34, “the Charter”), there are current AI eligibility assessment practices that have damaged citizens’ social rights. Acoording to Le Cuadernique du Net (2022), France’s Caisse des Allocations Familiale (CAF) claims that around 100,000 citizens have been wrongly denied benefits since 2021. Brazil’s Instituto Na- cional de Seguridade Social (INSS, 2024) says that automated analysis increased from 17% to 23% between 2022 and 2023, and that rejected social security claims increased from 50% to 70% since the introduction of AI systems for benefit eligibility. And the United Kingdom (UK) uses an AI system that has been overestimating people’s income. All these initiatives impact directly in citizens’ lives as social security is tied to vulnerability criteria, such as un- employment and employment injuries and preventing citizens from starving or giving them bare minimum life conditions (Maxwell, 2021). We expect to examine and contribute to the regulatory framework for AI through data and legal comparison methods, as one of the next main steps in AI implementation is to understand its legal categories (in this case, high risk and social scoring) through empirical data and already ongoing uses of AI systems, such as AI assessment of eligibility for social security benefits. 2. FRAMING THE SYSTEMS FOR RECOGNISING SOCIAL PROTECTION RI- GHTS IN THE EUROPEAN AI REGULATION Risks are inherent to work and life, to human beings themselves. Social protection against these risks, as a public policy, appears to be a way of holding together the social fabric, be-