With the rapid development of Large Language Models (LLMs), Knowledge Distillation (KD) has demonstrated significant potential in reducing model deployment costs. However, its application has raised substantial challenges concerning trade secret protection among LLMs. This study examines the legal regulation of KD through a comparative analysis of major jurisdictions, including the United States, the European Union, Japan, and China. The research reveals that KD has transformed legal disputes from traditional relationships into Business-to-Business (B2B) relationships, thereby presenting novel challenges to trade secret protection systems. To address these challenges, this study establishes a dynamic theory at the theoretical level, proposes “forward licensing” models and staged compensation mechanisms at the institutional level, and constructs a multi-dimensional evidence collection system at the practical level, encompassing functional feature comparison and performance curve analysis. This research not only expands the theoretical boundaries of trade secret but also provides actionable legal tools for regulating emerging technological competition.