AI regulatory sandboxes as legal transplants: governance, regulatory learning and legal-technical interaction

This article analyses the AI regulatory sandboxes architecture under the AI Act through the lenses of “legal transplantation”, advocating for a multilevel examination of their legal, institutional, and functional implications. AI regulatory sandboxes are conceived as controlled spaces for developing, training, validating, and testing AI systems subject to regulatory supervision. The article explores three key dimensions for their implementation, as key factors for a successful “transplant”: governance, regulatory learning, and legal-technical interaction. Thus it first examines multilevel coordination problems at the interface between EU institutions, national and sub-national governments, and sectoral regulators, supporting harmonisation and structures of accountability. Then, it addresses the AI regulatory sandbox as a regulatory learning instrument, through which competent authorities may adapt not only the applicable rules but also their practices and regimes in response to sandbox experimentation. Lastly, the article addresses the fundamental issue of “substantial modification” in AI systems and the role of AI regulatory sandboxes in testing and supporting its assessment.