Systematic review of Regulatory Sandboxes : implications for the European Union's Artificial Intelligence Act

Abstract

The UK’s Financial Conduct Authority (FCA) introduced the Regulatory Sandbox for the financial technology market 2016. Since then, many regulators worldwide have adopted and implemented the model based on their conditions. The European AI Act was first announced in 2021 and will enter into force in 2024. The AI Act is planned to go into effect in 2024 and, among others, provides definitions for high-risk AI systems and requirements for an AI regulatory sandbox, which motivates this research.

This research aims to conduct a systematic review of academic literature on regulatory sandboxes to understand and identify possibilities for their usage for AI systems based on the AI Act. The systematic review presents a comprehensive and structured understanding of regulatory sandboxes, dividing them into three distinct stages. The first stage includes the application process and eligibility criteria. The second stage focuses on the duration of a firm’s testing of its innovative product within a regulatory sandbox, conducted in a limited real-world environment with real clients under the supervision of regulators. The third stage explores the process of a firm exiting from a regulatory sandbox and the subsequent impact.

Based on the lessons learned from the SLR process, these findings present practical implications for understanding AI systems and their position within society compared to Fintech services. It also suggests a regulatory body for systematically comprehending and regulating an AI system, providing a roadmap for effective AI governance.