Who Foots the Bill? State-Backed No-Fault Compensation for Experimental Harms in Artificial Intelligence Regulatory Sandboxes

Abstract

As emerging technologies such as artificial intelligence (AI) outpace existing legal frameworks, regulatory sandboxes have become a key mechanism to reconcile innovation with risk governance by allowing firms to test novel technologies under regulatory oversight with certain temporary legal exemptions. While effective in fostering innovation, these frameworks leave unresolved liability concerns when harm results from compliant experimentation. Existing models place the burden of compensation primarily on participating firms, neglecting the potential role of state-backed liability mechanisms. This paper critically examines the accountability gap in regulatory sandboxes for AI by analyzing the limitations of current compensation frameworks used in leading jurisdictions. It argues that existing sandbox models, which primarily allocate civil and criminal liability to participating firms, are ill-equipped to address the distinct harms posed by AI technologies. These inadequacies arise from legal exemptions that undermine access to fault-based remedies, the technical opacity and distributed agency of AI systems, and the systemic nature of many AI-related harms. To address these challenges, the paper proposes a state-backed no-fault compensation scheme, modeled on Vaccine Injury Compensation Programs (VICPs), to more equitably and effectively distribute liability. This paper argues that the state’s active involvement in sandbox governance renders it a co-responsible actor, warranting shared liability in harm cases. A state-backed no-fault approach would also help overcome the attribution challenges posed by opaque, distributed AI systems, reduce barriers for small and medium-sized enterprises (SMEs) in low- and middle-income countries (LMICs), and strengthen public trust in experimental regulatory regimes.