摘要：

目的 以“治理工具组合”为分析框架，量化刻画我国学术期刊生成式AI披露治理的规范可实施性，并进一步检验期刊治理预期与作者披露行为实践之间的落差，以期为期刊编辑部与学术治理机构优化生成式AI规范体系、提升治理落地效能提供量化依据与决策参考。 方法 在治理端，对我国学术期刊AI治理文本按意义单元进行编码，构建透明度产出指数（transparency output index，TOI）与可执行性指数（executability index，EI）并归一化，引入文本可见性指标衡量对作者的可触达性；在作者端，在问卷中结合情境实验，比较“惩罚明确性”与“文本可见性”对作者披露预期的影响。 结果 治理侧指标显示，62%的期刊缺口指标（Gap）>0（均值0.133），透明度产出优于可执行性；情境实验表明，惩罚后果明确能显著提升作者披露预期（OR≈2.05，p=0.020），而单纯提高文本可见性效果不显著（OR≈0.80，p=0.505）。 结论 建议从审稿环节入手，明确编辑、审稿人在内的主体责任归属；同时，将工具名称、使用环节、使用范围和人工核对声明等最低披露信息项嵌入投稿系统必填字段，并配套抽查机制和分级处置规则形成核验闭环，以提高披露要求的可实施性。在行业层面，需要协同推进披露模板、系统模块、培训和争议处理机制。

关键词: 生成式AI披露治理, 学术披露, 学术规范, 制度设计, 治理评估

Abstract:

Purposes Using the governance instrument mix as the analytical framework， this study quantitatively examines the transparency output and executability of generative AI governance among Chinese academic journals. It further assesses the gap between journal-side governance expectations and authors’ disclosure practices， with the aim of providing empirical evidence for improving the implementation of generative AI policies in academic publishing. Methods On the governance side， AI-related policy texts from Chinese academic journals were segmented into meaning units and coded to construct and normalize the transparency output index （TOI） and executability index （EI）. A policy visibility index was also introduced to measure the accessibility of governance texts to authors. On the author side， a survey with embedded vignette experiments was used to compare the effects of penalty clarity and policy visibility on authors’ disclosure intentions. Findings Governance-side indicators show that 62% of journals have a positive gap index （Gap > 0； mean=0.133）， suggesting that transparency output generally exceeds executability. The vignette experiments further show that explicit penalty consequences significantly increase authors’ disclosure intentions （OR≈2.05， p=0.020）， whereas improving policy visibility alone has no significant effect （OR≈0.80， p=0.505）. Conclusions The findings suggest that generative AI governance should be strengthened from the peer-review workflow by clarifying the responsibilities of editors， reviewers， and other relevant actors. Journals should embed a minimum set of disclosure items into mandatory submission-system fields， covering tool identification， use context， and human verification， supported by audit mechanisms and tiered response rules. At the industry level， coordinated efforts are needed to develop standardized disclosure templates， system modules， training programs， and dispute-resolution mechanisms.

Key words: Generative AI disclosure governance, Scholarly disclosure, Academic norms, Institutional design, Governance assessment