生成式人工智能训练未经著作权人许可使用海量受版权保护的作品，在我国现行法律制度下需要承担侵权责任，版权合规成为从业者必须面对的重要课题。然而，传统“授权后利用”的合规方式在人工智能领域窒碍难行，合理使用规则又无法豁免营利性模型训练，探索新型合规路径有其紧迫性与必要性。应当从具有终局性的司法裁判出发，明确当前我国对于人工智能训练版权问题的裁判思路，倒推出具体可操作的合规路径，为人工智能产业发展提供指引。

When generative artificial intelligence uses a vast amount of copyrighted works without the copyright holders’ permission during the training process, it is liable for infringement under China's existing legal framework. Copyright compliance has thus emerged as a crucial challenge that industry players cannot afford to ignore. However, the conventional compliance model of “authorization before utilization” encounters significant obstacles in the artificial intelligence domain. Moreover, the fair use doctrine fails to exempt profit - making model training, rendering the exploration of novel compliance approaches both urgent and indispensable. By delving into conclusive judicial decisions, we can discern the current judicial stance on copyright issues in artificial intelligence training in China. From this, we can reverse-engineer practical and actionable compliance strategies, thereby offering valuable guidance for the development of the artificial intelligence industry.