THE AI ACT: REGULATORY AND TECHNICAL CHALLENGES

There is little doubt that the AI Act will remain one of the most influential pieces of Artificial Intelligence legislation in the coming years. This is due both to its broad geographical and material scope and to the fact that it was developed slightly before the worldwide boost in regulatory efforts, thus rapidly becoming a reference point. Hence, my Thesis proposes to analyse the AI Act, its benefits and drawbacks in the level of protection of fundamental rights through the analysis of its structure and regulatory approaches, also with a view to the broader AI governance and regulation framework. In fact, by regulating most economic sectors with comprehensive provisions, the AI Act shows difficulties in terms of resilience and flexibility. These regulatory trade-offs and challenges are discussed in the Thesis both at a higher and at a more granular level, with a focus on five (four plus one) legal institutions. As for the latter, firstly, the Thesis addresses legal remedies to technological limitations in selected rights-intensive provisions, analysing the concepts of transparency and bias. Second, it discusses risk management and harmonized standards (i.e. technical remedies to regulatory limitations) as system-closing rules necessary to ensure that the regulatory architecture is applicable and enforceable. Third, the regulation of general-purpose AI will serve as a stress test for the stability of the AI Act’s structure and for the protection of fundamental rights. Ultimately, my work aims to support efforts to analyse such a complex and lengthy regulation, with particular consideration to the operationalization of fundamental rights, which is complicated by the regulatory structure of the AI Act, as mentioned above. More than that, this Thesis strives to show that the proper interpretation and compliance with regulations such as the AI Act should not hinder innovation but channel it towards applications that do not only consider commercial interests but are also human-centred.