From Simple Sandbox Process to Regulatory Sandbox Framework: Serving the Dual Objectives of AI Regulation

Abstract

In this paper, we build on a simplified process to evaluate AI systems compliance with AI Act. As AI Act’s risk-based regulatory measures simultaneously strive for providing innovative, but trustworthy AI systems to the market, we extend the process to an AI Regulatory Sandbox Framework (AIRSF), which is connected with the EU’s innovation ecosystem. The study builds on a design science approach combined with action research sandbox as a boundary object for multistakeholder sandbox development. We first analyse the options for regulating AI Systems. In doing so, we explore the regulatory sandbox as a means aiming to remedy the known limitations of regulatory sandboxes, especially from the viewpoint for SMEs . We conclude by reflecting upon the findings during the sandbox development process.