New Tools for Experimental(ist) Governance: the Case of AI Regulatory Sandboxes in the EU

Regulatory sandboxes have emerged as a staple of AI governance strategies, appearing repeatedly in national AI roadmaps as flexible mechanisms for keeping pace with rapid technological advances. The European Union has elevated this mechanism, introducing the most comprehensive framework to date that that makes regulatory sandboxes for AI systems mandatory in all member states. This paper analyses the newly introduced administrative and governance complexity for the regulators (who have to include a variety of actors, such as all the AI factories, hubs, etc and guarantee confidentiality and transparency), drawing on the draft Implementing Act issued by the European Commission and the responses received to the consultation over 6 weeks. Using qualitative methodology and thematic analysis, we position the choices in the Implementing Act against the broader objectives of regulatory sandboxes, from regulatory learning to societal benefits. We also highlight the convergences and divergences with other regulations, showing how the EU is developing its experimentalist approach.