|
Abstract The rapid development of generative AI has brought both technological revolution and multiple security risks. How to balance risk regulation and innovation incentives has become an important issue. Regulatory sandbox, as an innovative regulatory tool, responds to the requirements of inclusive and prudential regulation and the concept of agile governance. However, domestic research on regulatory sandboxes mainly focuses on the field of Fintech, and there is no research on regulatory sandbox for AI, especially for generative AI. This study aims at exploring the effective regulation and innovation incentives for generative AI through the construction of the regulatory sandbox for generative AI.
Existing regulatory models, whether “command-and-control” or “development first, governance later”, have difficulties. The entrance of inclusive and prudential regulation and the concept of agile governance has become inevitable. Based on it, regulatory sandboxes for generative AI have feasibility in China. The regulatory sandbox for AI has been applied in more and more extra-territorial countries. Some countries are also exploring the construction of the regulatory sandbox for generative AI. The domestic regulatory sandbox has also accumulated a certain amount of experience, and the generative AI industry is also accelerating. It should be noted that AI involves innovative projects in different scenarios and fields, which means that a single, undifferentiated regulatory sandbox for AI may be more inappropriate. What’s more, the expression “regulatory sandbox” is more appropriate for regulatory sandbox for generative AI. In terms of governance effects, the construction of a generative AI regulatory sandbox is beneficial to multiple actors, reducing risks for innovators, improving regulation, and protecting consumers.
The path to implementation of regulatory sandbox for generative AI is discussed next. At the macro level, two schemes are matter of expediency. One is to authorize the regulatory sandbox through the regulatory documents of the State Council or relevant ministries, and gradually refine it at the local level, and the other is to incorporate the provisions of the regulatory sandbox into “Interim Measures for the Administration of Generative Artificial Intelligence Services”. Instead, the following two options should be adopted to make the regulatory sandbox legally enforceable. One is to make a decision by the National People’s Congress and its Standing Committee, and the second is to formulate a unified Artificial Intelligence law and write it into the regulatory sandbox. At the micro level, the entire regulatory process should be designed. First, at the application stage, the entry threshold should be determined and three categories and six principles should be constructed, namely, “substantial innovation”, “tilting towards small and micro enterprises”, and “risk regulation needs”, “project maturity”, “public interest principle” and “domestic market application principle”. Secondly, in the testing phase, clear exemption rules and effective communication are required to build exemption rules based on time, application scenarios and user types, as well as a mechanism for disclosure of information during the whole process and effective communication between regulators and innovators. Third, in the exit phase, planning the exit conditions, regulators need to make two types of five judgement results for different situations; after exit, regulators should require innovators to complete their exit obligations, properly handle data and personal information, and jointly prepare regulatory sandbox exit reports. It is important that we continue to explore and improve the regulatory sandbox for generative AI in practice so that generative AI can better promote China’s economic and social development.
|
|
Published: 22 March 2026
|
|
|
|
