An academic journal that published a paper containing fictitious AI-generated references has said the work’s core conclusions remained valid despite “some mismatches and inaccuracies” with the citations, following an investigation launched by the University of Hong Kong, where the author is studying for her PhD.
Highlighting concerns raised by this case, veteran journal editor Robert Davison called for better enforceable policies to regulate AI’s role in academic work.
Davison, a professor of information systems at City University, added that while major journals typically had their own policies on the use of artificial intelligence (AI) for academic work, enforcement ultimately depends on editors, who must carefully verify reference lists.
In the paper, titled “Forty years of fertility transition in Hong Kong” and published in China Population and Development Studies on October 17, at least 20 out of 61 references appear to be non-existent.
The paper’s corresponding author, Professor Paul Yip Siu-fai of the social work and social administration department at the University of Hong Kong (HKU), apologised on Sunday on behalf of himself and his PhD student, Bai Yiming.
The allegation first surfaced on social media after a user claimed she “was told by a friend” that many of the paper’s references appeared to be the results of “AI hallucinations”.
Yip told the media that Bai, the lead author, had used AI to help with referencing but failed to verify the citations. Yip, who served as the corresponding author responsible for oversight, said he found out about it from Bai after seeing the allegations online.
He said the issue did not constitute a breach of academic integrity as the content of the paper, which was accepted after two rounds of academic review, was not fabricated.
The paper’s other authors were involved in giving advice and helping with data analysis, he said.
He also said he had notified the journal and would submit the correct version in a few days, pledging to scrutinise every word and ensure it is error-free.
Yip, an associate dean of the social sciences faculty and the chair professor of population health, is a member of the editorial board of the journal.
China Population and Development Studies received its first official journal impact factor of 1.1 in June 2025. The index is a measure of the frequency with which the average article in a journal has been cited in a particular year.
The journal’s editorial office said the author reported “some mismatches and inaccuracies in the reference list and corresponding in-text citations, which were due to their oversight before submission”.
It said the core conclusion, empirical data validity, and theoretical framework integrity were unaffected, and confirmed the core argument chain was independent of reference issues.
A supplementary correction would be released on the website as soon as possible, it said.
According to the HKU website, Bai is a PhD candidate with a research focus on youth sexuality, population health and policy analysis. She contributed six articles to the Post between 2023 and 2024.
The paper’s other authors include Billy Li of the government’s Census and Statistics Department, Dr Wat Kam-pui of the statistics and actuarial science department at HKU, Dr Eddy Lam of the nursing and health sciences school at Hong Kong Metropolitan University, and Dr So Bing-kwan of the mathematics school at Jilin University in mainland China.
The Post had reached out to them all, but none of them replied by the deadline.
At least 20 references appear to be fictitious in the paper. For example, the first citation on the list – “The changing role of women in the family: Implications for fertility” – said to have been published in the Journal of Marriage and Family in 2000, does not exist based on a search by the Post on Google Scholar.
Some of those references were cited with a digital object identifier or a link to Google Scholar, but they led nowhere.
Notably, one citation even referred to a non-existent paper which listed Yip as the author.
According to Springer Nature, the publisher of the journal, its editorial policies prohibit crediting large language models as authors, and require that any use of such tools must be disclosed except when limited to basic copy editing, such as grammar and style.
Peer reviewers are also barred from uploading manuscripts onto generative AI tools due to the risk of inaccuracy and breaches of confidentiality.
Any AI tool used to support the report must be transparently declared.
However, the paper in question contained no such declaration.
HKU said it had a strict set of policies and guidelines governing the use of AI in academic research and has initiated a formal investigation.
It said disciplinary action would be taken if any breaches of research ethics and standards were identified.
Davison, who edits the Information Systems Journal published by Wiley-Blackwell, said he permitted AI use only for non-intellectual tasks such as grammar and spelling checks, but not for generating content or referencing.
Wiley’s guidelines stipulate that if an author has used any AI tool to develop any portion of the manuscript, its use must be described in detail. Whatever the tool provides, the author is responsible for its accuracy and for listing the correct references.
The final decision about whether the use of such a tool is appropriate or permissible in a certain paper lies with the journal’s editor or other party responsible for the editorial policy.
At his journal, each paper typically has about five editors and reviewers.
Davison said he did not need to check every reference individually because he was already familiar with many of the authors in the field and could quickly spot fake citations, some of which were even attributed to him.
He added that when such cases arose, he would immediately reject the paper and if the case was particularly severe or egregious, inform the authors’ institutions.
However, he believed there was no consensus across disciplines and their editors in different countries on how serious the misuse of AI in academic work should be considered, which warranted more communication among editors.
Davison warned that the problem could worsen and become “normalised” if there were no clear, enforceable policies and proper education to regulate AI’s role in academic work, which should be the responsibility of journal editors and publishers.
The Post has reached out to Springer Nature and the journal for an explanation of the oversight. - SOUTH CHINA MORNING POST
