Abstract: Currently, while artificial intelligence is driving social transformation, it also brings numerous risks and challenges, highlighting the "Collingridge Dilemma". In response, China has been exploring artificial intelligence(AI) sandbox regulation through regional and sector-specific pilot programs. The sandbox regulatory approach, originally applied in computing and later in fintech, has undergone a process of "transplantation" and "re-transplantation" into the field of AI. Its core objective is to contain risks within a limited scope through regulatory experimentation, addressing the dual uncertainties associated with emerging technologies while fostering their development. However, current practices reveal that China's AI sandbox regulation faces challenges at the procedural, managerial, and participatory levels, hindering its effectiveness. Procedurally, there is a lack of clear entry and exit criteria, as well as insufficient integration with conventional regulatory processes. Managerially, the flexibility and guidance of AI sandbox regulation remain weak. In terms of participation, the structure is predominantly binary, involving mainly regulators and regulated enterprises, with limited involvement of independent third parties such as experts and consumers. To address these issues, it is essential to first recognize that, given the layered industrial chain of AI and the current pause in specialized legislation, China should prioritize sector-specific approaches in advancing AI sandbox regulation. Secondly, procedurally, entry and exit standards should be refined across different levels and sectors, while ensuring effective linkage with external conventional regulatory procedures. Thirdly, managerially, a layered and categorized governance approach should be adopted, including the establishment of a sandbox specialist system. Finally, in terms of participation, consumers and independent third-party institutions should be incorporated into the sandbox regulatory framework to promote multi-stakeholder collaboration.