生成式AI训练数据的著作权侵权规制路径

Regulatory Approaches to Copyright Infringement in Generative AI Training Data

摘要: 生成式AI发展离不开海量高质量训练数据，其数据获取与使用常涉及受著作权保护的作品，引发产业发展与著作权保护的突出矛盾。现行授权许可、合理使用、法定许可等著作权规制路径存在实操困难、场景适配不足、机制僵化等局限，难以应对相关问题。准法定许可制度通过公告异议、默示同意、合理付酬的设计，兼顾AI开发者授权成本与著作权人合法权益，契合知识产权制度本质与国家发展战略，且有立法先例与技术支撑，具备可行性。构建该制度需明确权利义务、规范程序、完善配套，实现技术创新与权利保护的平衡，为生成式AI产业健康发展提供法治保障。

Abstract: The development of generative AI is inextricably linked to massive volumes of high-quality training data. However, the acquisition and utilization of such data frequently involve copyright-protected works, triggering a prominent conflict between industry development and copyright protection. Current copyright regulatory frameworks—such as authorized licensing, fair use, and statutory licensing—suffer from limitations including practical operational difficulties, insufficient adaptability to varying scenarios, and rigid mechanisms, making them inadequate for addressing these emerging challenges. A quasi-statutory licensing system, designed around mechanisms of public notice and objection, implied consent, and reasonable remuneration, effectively balances the licensing costs for AI developers with the legitimate rights and interests of copyright owners. This approach aligns with the fundamental principles of the intellectual property system and national development strategies. Supported by legislative precedents and technical feasibility, it represents a viable solution. Establishing this system requires clarifying rights and obligations, standardizing procedures, and perfecting supporting mechanisms to achieve a balance between technological innovation and rights protection, thereby providing robust legal safeguards for the healthy development of the generative AI industry.