With the expansion of artificial intelligence worldwide, countries have started adopting the sandbox regime commonly used in the financial sector, and considering the lessons accumulated in its application, to examine AI applications in various domains and contexts. The European Union has gone even further, with its new Artificial Intelligence Act obliging EU Member States to establish AI sandboxes. The purpose of this article is to propose an optimal model for regulatory cooperation in AI sandboxes at the national level. Drawing on the experience of Fintech sandboxes, including the Digital Securities Sandbox in the United Kingdom, the article examines various potential models for operating regulatory sandboxes in the field of AI: jointly administered cross-sector sandboxes, separate sandboxes with a single-entry mechanism, and separate sandboxes with independent entry points. Additionally, the article evaluates the mechanism of a joint regulatory committee, whether as a statutory body with decision-making power or as a coordinating committee based on soft law instruments such as memoranda of understanding. The article concludes that a single-entry mechanism is better suited for jointly administered cross-sector sandboxes than for multiple sandboxes managed separately by different regulators. Furthermore, it finds that soft law instruments are preferable to hard law due to the dynamic nature of artificial intelligence and the need for flexibility, adaptability, and swift regulatory responses.

Read the entire article here: https://ssrn.com/abstract=5199887 or http://dx.doi.org/10.2139/ssrn.5199887