Abstract: The application of generative artificial intelligence (GenAI) technology in scientific journal publishing has extended from the explicit stage of paper writing to the relatively implicit process of peer review. Although editorial departments of journals both domestically and internationally have generally issued relevant usage regulations, the phenomenon of "clandestine use" to circumvent supervision during the review process persists, posing practical challenges to review confidentiality, originality of comments, and accountability. Based on the management practices of scientific journal publishing and analyses of public policy texts, this paper systematically examines the common dilemmas faced by two mainstream policies—“absolute prohibition” and “restricted disclosure”—during implementation. The study suggests that the insufficiency of policy effectiveness primarily stems from a gap between the editorial department's management perspective and the behavioral logic of reviewers: reviewers are driven by practical needs such as efficiency enhancement and knowledge supplementation, coupled with a low perception of violation costs; meanwhile, editorial departments face actual constraints in technical verification, resource allocation, and confidentiality risk prevention. Relying solely on prohibitions or disclosure requirements is insufficient to form effective constraints. Therefore, this paper proposes shifting the governance approach from “unidirectional constraint” to “collaborative guidance”. Editorial departments can adopt practical measures such as formulating detailed operational guidelines, embedding ethical clauses into the review process, developing and promoting secure localized auxiliary tools, and exploring the establishment of a credit incentive system for reviewers. By understanding the legitimate needs of reviewers, editorial departments can guide individual behaviors to align with the journal’s quality assurance objectives, ultimately fostering a responsible review environment that effectively mitigates risks while reasonably enhancing efficiency.