La matrioska della finanza: Un approccio multilivello per la governance dell’intelligenza artificiale nei mercati finanziari

This contribution adopts a multidisciplinary perspective to analyze the multiple risk profiles and systemic distortions associated with the introduction of artificial intelligence in the financial sector. Starting from the observation of the growing centrality of artificial intelligence in credit assessment, risk management, fraud prevention, and the personalization of financial services, the paper highlights how these technologies, alongside significant opportunities in terms of efficiency, inclusion, and innovation, also raise substantial concerns with regard to transparency, fairness, security, and the protection of fundamental rights. The current European regulatory framework, although grounded in a pragmatic and risk-based approach, appears fragmented and structurally ill-suited to govern risks that are inherently dynamic, endogenous, and mutable over time. In light of the emerging and persisting criticalities, the contribution argues for a rethinking of the role of supervisory authorities, which should operate not merely as guarantors of formal compliance, but as central actors within a more dynamic and layered model of financial governance. Such a model should be capable of integrating research activities, regulatory experimentation, continuous monitoring, and the use of algorithmic supervisory technologies. By drawing on forms of cross-supervision, the paper outlines new approaches to the management and co-construction of computational checks and balances, designed to detect anomalies, biases, and systemic risks in a timely manner, without resorting to automatic prohibitions or excessively restrictive solutions. In this perspective, the contribution situates itself within the broader debate on the limitations of traditional models of technological regulation, advocating an evolution towards dynamic, participatory, and lifecycle-oriented forms of regulation for artificial intelligence systems, aimed at ens