Abstract: Traditional regulatory models have become increasingly inadequate in addressing the complexity and rapid evolution of artificial intelligence (AI) technologies. Regulatory sandbox have gradually expanded into the AI domain, emerging as tools to balance security and development by offering flexible regulatory spaces.These sandboxes have been adopted by numerous countries worldwide.At present, mechanism design of the AI regulatory remains in the exploratory stage, with divergent practices across countries in key aspects such as legal authorization, regulatory boundaries, and governance models.This paper reviews the origin, evolutionary path and international application of AI regulatory sandboxes, and adopts the framework of "experimentalist governance" to guide the construction and operation of regulatory sandboxes.We summarize the operation processes of artificial intelligence regulatory sandboxes in 9 countries around the world, and extract the mechanisms of framework goal setting, independent experimental execution, supervision and feedback, evaluation and adjustment.This mechanism illustrates the applicability of regulatory sandboxes in enabling dynamic policy adjustment, fostering multi-stakeholder collaboration, and promoting regulatory innovation. It facilitates the continuous optimization of policy objectives by encouraging effective interactions among regulators, innovators, and consumers through real-time feedback and coordinated efforts, thereby fostering a positive relationship between technological innovation and regulatory compliance.The development of AI regulatory sandboxes should focus on sector-specific regulatory approaches, the formulation of measurable application criteria, and the enhancement of international cooperation and mutual recognition mechanisms to address global governance challenges posed by AI technologies. By promoting the localized evolution of sandbox systems and actively participating in the formulation of international standards, China can enhance its AI governance capacity between domestic and international regulatory systems.

摘要： 传统监管模式已难以有效应对人工智能技术所带来的变化。源于金融科技领域的监管沙盒扩展至人工智能领域，成为平衡安全与发展的工具，为人工智能技术提供灵活的监管空间，已被全球多个国家所采纳。当前，人工智能监管沙盒的机制设计尚处于探索阶段，在法律授权、规制边界及治理机制等关键环节上，各国实践表现出路径分化。本文通过梳理人工智能监管沙盒的起源、演进路径及国际应用实践，以“实验主义治理”作为指导人工智能监管沙盒构建与运行的理论框架，比较全球9个国家人工智能监管沙盒运行流程，提炼出框架目标设定、实验自主执行、监督与反馈、评议与调整四项机制，解释监管沙盒在推动动态调整、多方协作与政策创新方面的适用性。监管沙盒通过对政策目标的持续优化，推动监管主体、创新主体与消费主体在实际运行中通过多方协作和实时动态的反馈渠道优化监管框架，实现技术创新与合规之间的良性互动。发展人工智能监管沙盒可从推进分领域监管沙盒体系化布局、建立可量化的申请标准与评分机制、构建协同公职的合作机制架构、探索跨国互认与标准协调机制着手，提升中国人工智能治理能力，促进国内外监管体系的协同与创新，应对全球技术带来的治理挑战。