生成式人工智能数据训练的版权侵权风险应对研究

A Study on Addressing the Copyright Infringement Risks in the Data Training of Generative Artificial Intelligence

摘要: 在生成式人工智能发展的过程中，数据训练对最终生成结果的效果起到了关键作用。现阶段数据训练未经授权而使用海量受保护的作品，版权侵权风险显而易见，而学界及业界并未就数据训练的法律定位形成一致观点。通过对学界观点的分析比较和对域外立法经验的借鉴，在我国合理使用制度框架下提出构建数据训练的合理使用规则和相应的配套措施，能够在兼顾著作权人利益的同时保障人工智能产业的发展，化解数据训练阶段的版权侵权风险。

Abstract: The development of generative artificial intelligence (AI) critically depends on large-scale data training, which fundamentally determines the efficacy of the outputs. However, the prevalent practice of utilizing massive amounts of copyrighted works without authorization for training purposes presents evident infringement risks. Currently, there is no consensus within either academia or industry regarding the legal status of such data training activities. This study addresses this issue through a comparative analysis of scholarly perspectives and an examination of extraterritorial legislative approaches. It proposes a framework within China’s existing fair use system to mitigate copyright risks at the data-training stage. The core argument is that establishing tailored fair use rules, supplemented by corresponding supporting measures, can effectively reconcile the interests of copyright holders with the developmental needs of the AI industry. This approach aims to foster innovation while providing a legal pathway to resolve the inherent copyright infringement risks associated with data training.