Regulatory Sandboxes for Artificial Intelligence: Reconciling Anticipatory Regulation and Legal Certainty

In late March 2023, ChatGPT made headlines when the Italian Data Protection Authority (the Garante) issued a temporary ban on the service due to concerns related to the processing of personal data. The regulator’s reactive approach demonstrated the recurring challenge to apply traditional regulatory approaches to fast-emerging disruptive technologies such as artificial intelligence (AI). The struggle for change in the regulatory approach became prevalent in the wake of the 2008 financial crisis, and subsequently led to the development of new tools of regulation, such as regulatory sandboxes. While the scope of the term remains subject of discussion, the EU recognised the potential of the novel anticipatory, as opposed to reactive, regulatory method by recognising it normatively in the proposal for the AI Act. The starting point of the proposal is the systematic exploration of the evolution of regulatory sandboxes and their place in contemporary regulatory theory in order to better understand their positive and negative effect on AI governance. This enables further evaluation of how such instruments could impact legal certainty as a pivotal element of the rule of law. To this end, this project traces the evolution of the understanding of legal certainty in the EU by combining doctrinal and empirical research methods. The final goal of the thesis is to propose normative recommendations for the design and implementation of regulatory sandboxes for AI.