Getting Regulatory Sandboxes Right: Design and Governance Under the AI Act

Abstract Regulating emerging technologies involves balancing the mitigation of risks with the promotion of innovation; a balance frequently seen as a zero-sum “dilemma of control.” Regulatory sandboxes offer a practical way to address this dilemma by enabling controlled, evidence-based testing of new technologies. In this article, we examine the regulatory sandbox framework introduced by the EU Artificial Intelligence Act (AIA). We argue that the AIA’s multi-level governance structure represents a shift from traditional sandbox models by prioritising regulatory learning over technological disruption and expanding public interest considerations to include strategically aligned commercial innovations. Afterwards, we identify governance challenges across three sandbox phases – pre-testing, testing and post-testing – and propose structured solutions. Our analysis suggests that effective sandbox governance requires specific mechanisms: tailored entry criteria, precise pipeline placement guidance and multi-agency coordination in pre-testing; experimental realism and continuous risk classification updates during testing and clear graduation criteria with robust transition support in post-testing.