Descriere
This study analyses the technical and legal challenges of implementing the right to erasure provided for in Article 17 of Regulation (EU) 2016/679 in the context of voice cloning systems based on artificial intelligence. The central hypothesis is that, although the architecture of machine learning models raises significant technical obstacles to selective data erasure, the obligation of operators to ensure the effective exercise of this right remains fully applicable. The objective of the research is to identify and legally assess emerging technical solutions for compliance with the GDPR requirements. The methodology combines normative analysis with the examination of technical solutions such as machine unlearning, model marking and logical deactivation. The results confirm that these techniques offer viable paths to compliance, subject to the fulfilment of criteria of irreversibility, inaccessibility and demonstrability. The practical implications concern the need to adopt specific guidelines and integrate erasure facilities from the design phase of systems.
