Recently, the research paper titled "AI Policies in Academic Journals Fail to Curb the Surge in AI-Assisted Academic Writing" was accepted by the Proceedings of the National Academy of Sciences (PNAS). Authored by He Yongyuan (sole first author), an undergraduate student from the 2023 cohort of the Big Data Management and Application program at Peking University’s Department of Information Management, and Assistant Professor Bu Yi (sole corresponding author), this study represents one of the key outcomes of a project funded by the National Natural Science Foundation of China and led by Bu Yi.

With the widespread adoption of generative AI, significant shifts have occurred in the landscape of academic writing. While these tools have enhanced research efficiency, they have also sparked debates over issues such as content “hallucinations,” ethical risks, and the proper attribution of AI contributions to authorship. In response, publishers and academic journals globally have introduced policies aimed at safeguarding research integrity. However, the actual effectiveness of these measures remains uncertain.

Through large-scale empirical analysis, this paper evaluates the real-world impact of AI usage guidelines on academic writing practices and transparency. The study examined 5,114 journals from the JCR Q1 quartile and 5.2 million papers published between January 2021 and June 2025, conducting in-depth analysis of 164,000 full-text papers. The findings reveal that despite 70% of journals having established AI policies, the use of AI writing tools among researchers has risen sharply, with no significant difference observed between journals with and without such policies. Notably, AI tool adoption has grown significantly in non-English-speaking countries, natural science fields, and high open-access journals.

Full-text analysis indicates that since 2023, only about 0.1% of published papers explicitly disclose AI usage—a stark contrast to the actual high rates of AI tool adoption. The study underscores that current policies have not effectively regulated AI usage but have instead guided researchers toward shifting from mere “usage” to “declaration,” promoting greater transparency in academic practices.

The paper calls on the academic community to reevaluate existing ethical frameworks and transition toward more inclusive and practical governance models. Such efforts are essential to foster responsible integration of AI in human-machine collaborative research paradigms.