| Using 89 generative AI statements from Chinese academic journals (2023–2025), this study employs LDA topic modeling to analyze their thematic structures, evolution, and governance logic. Five core themes emerge: "academic norms," "use disclosure," "review mechanisms," "accountability," and "academic integrity." Thematic evolution unfolds in three stages: S1 relied on declaratory approaches, focusing on ethical initiatives and discussions of legitimacy; S2 shifted to procedural regulation with review and disclosure frameworks; S3 advanced to punitive governance, centered on quantitative accountability and boundary-setting. Despite growing regulatory rigor, challenges persist—fragmented consensus, enforcement doubts, and replication difficulties. The study recommends moving beyond technological detection, establishing humans as sole responsibility holders, building transparent mechanisms grounded in procedural justice, and developing disciplinary alliances to create unified standards—transitioning from journal-specific policies to community-wide norms. |
