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.