Abstract
Respected Editor,
A few days ago, we received an expert review of our systematic review and meta-analysis sent to the journal for possible consideration toward publication. One of the reviewers provided detailed feedback across multiple sections and recommended that we cite three additional systematic reviews that would strengthen our introduction and discussion. Each recommendation came with complete bibliographic details, including author names, journal titles, publication years, and DOI numbers. As conscientious researchers, we began revising our manuscript to incorporate the reviewer’s suggestions. Until we tried to verify the citations.
The Discovery
A careful search of multiple academic databases (PubMed, Scopus, Web of Science, and Google Scholar) showed striking discrepancies. The first study could not be located in the literature, while its DOI belonged to another article. Both the second DOI and the associated article did not exist. The third reference, while real, was not directly relevant to the scope or methodology of our study. Artificial intelligence (AI) language models are known to generate convincing but fictitious bibliographic information, a phenomenon known as “hallucination”1, 2. If a reviewer used such tools without verification, it could explain the specific error pattern we observed. A detailed perusal of the remaining comments revealed telltale indicators of automated generation, which were submitted without adequate oversight or disclosure by the reviewer.
The Implications
Fabricated citations in a peer review point toward reviewer misconduct, a phenomenon with wide-reaching repercussions that pose fundamental threats to the integrity of the scientific community, enlisted as under:
Degradation of the scientific record: Generative AI inherently lacks understanding of domain-specific knowledge, methodological intricacies, and contextual relevance critical to the evaluation of a manuscript. 3 Maintaining factual precision is the footing on which the entire scientific process stands tall, and this should be safeguarded diligently. Uploading manuscripts onto generative AI platforms compromises confidentiality and potentially violates copyright. 4
Erosion of trust: The peer review system presumes that reviewer recommendations are based on expert evaluation. Submission of AI-generated content by a reviewer without meticulous verification violates the trust that editors and authors repose in the peer review system. 5
Systemic vulnerability: This incident raises a doubt as to how many other such instances have escaped detection. How many authors or reviewers have unwittingly (or worse, willingly) integrated fictitious work into their research work without verification?
The Path Forward
It is imperative to emphasize that this letter is not a denunciation of any individual reviewer or the entire peer review process. Rather, it is a reflection on an acute and emerging concern in academic publishing: the unregulated use of AI in peer review, and the risks posed when such tools are used as a substitute for scholarly discernment rather than as an aid to it. The increasing use of AI tools for academic research and writing purposes is undeniable and inevitable. While much emphasis has been placed on authors’ roles and responsibilities with regard to the use of AI, their use in peer review calls for equally meritorious consideration and transparent regulations.6, 7 Some of the suggestions to tackle this threat include3, 4, 8:
Quality assurance protocols: Editorial boards should establish appropriate oversight mechanisms to detect use of AI, including verification of citations and cross-referencing of recommendations against established literature.
Reviewer training: Professional development programs should address the ethical implications of using AI for peer review and furnish guidance for responsible use of such instruments.
Transparency and mandatory disclosure: Disclosure of the use of AI tools by reviewers ought to be made mandatory, analogous to conflict-of-interest declarations, thus enabling editors to make informed decisions about the quality and reliability of the review.
Editorial guidelines: Publishers must stipulate clear policies with regard to the use of AI in peer review, specifically advising/mandating human verification of AI-generated content and directives for appropriate and inappropriate applications. Editors should be trained to recognize potential red flags in reviews, such as unusual citation patterns or recommendations that seem at odds from the manuscript’s primary focus.
Conclusion
It is in the nature of things that AI will be integrated into peer review, but the scientific community is obligated to ensure ethical, responsible, and transparent use of AI. Peer review remains, at its core, a human intellectual endeavor, requiring attention, care, and scholarly judgment. We hope this perspective contributes to a continuous dialogue on how to adapt our systems in response to emerging technologies without compromising scientific integrity.
We can and must do better.
