Commentary

Undeclared AI-Assisted Academic Writing as a Form of Research Misconduct

By: Bor Luen Tang
Tang BL. Undeclared AI-assisted academic writing as a form of research misconduct. Sci Ed. 2025;48.https://doi.org/10.36591/SE-4804-02
September 8, 2025
AI & Technology Authorship Early Online Ethics Peer Review

Abstract

The rise of generative artificial intelligence (AI) has broadened the academic editor’s job scope, and one must now worry about undeclared use of generative AI in manuscript preparation. While I agree that regulation of AI in academic publishing requires a tighter framework with better granularity beyond plain declaration, I posit that undeclared AI use is, in principle, a form of research misconduct on the part of the author. This is because such acts potentially constitute manipulation of the research process to produce unreliable, if not inaccurate, data/knowledge, which falls squarely under the research misconduct category of “falsification of research.”   

Introduction

Ever since the emergence of the large language model (LLM)–based AI chatbot ChatGPT in 2021, the use of generative AI (hereafter, AI) in academic writing has been somewhat controversial. While obviously immensely useful in many aspects of research, ChatGPT and many other LLM-based chatbots (henceforth, AI) subsequently introduced in quick succession have also been revolutionary in helping authors to write academic papers. Journals and publishers have quickly enacted rules on generative AI use in manuscript preparation, most of which mandate some form of declaration of such use (and the barring of AI coauthorship). However, several studies,1,2 as highlighted in a recent news article in Nature,3 have identified a small but significant number of papers with no declaration, but which clearly exhibited traces of AI writing. These findings, in addition to the rather eye-catchingly ridiculous figure of a rat with oversized genitalia,4 have deepened the feeling of unease with regards to illegitimate AI use in academic writing. 

Would the declaration of AI use suffice, and would it work to preserve integrity in academic publication? In the pages of Science Editor, Staiman5 has recently opined that mere mandates of declarations of AI use would not be sufficient and proposed a risk-based approach for better granularity in AI guidelines in publishing. On the other hand, Resnik and Hosseini6 have also proposed that such declarations be tiered as either mandatory, optional, or unnecessary depending on the nature of AI use. The situation is complicated for journal editors because of difficulties in ascertaining whether AI was indeed used in writing a manuscript, as AI detection software often does not have the necessary reliability.7 Should a manuscript with suspected undeclared AI be simply rejected or withdrawn, or might corrections and retrospective declarations be allowed? I take the view, which I will explain below, that undeclared (and particularly with veracity of content unverified) use of AI deserves no second chance.

The Wrong of (Undeclared) AI Use

What exactly is wrong with using AI to write manuscripts? Other than potential clashes with copyright laws,8 issues of research integrity would be a primary concern for an academic journal editor. In accordance with the widely adopted definition9 by the US Office of Research Integrity, misconduct in research involves either acts of fabrication, falsification, or plagiarism. The use of AI in academic writing has been questioned as to whether this constitutes a form of ghost writing or plagiarism,10 particularly when most parts of, or whole, manuscripts were generated by AI with little or no human input. I have previously argued that AI-based plagiarism, or AIgiarism, is a form of bypass plagiarism that risks propagation of erroneous and biased content.11 Shaw12 went further to state that misuse of generative AI is worse than plagiarism. Here, I posit that undeclared use of AI in academic writing when such declaration is mandated,13 potentially constitutes acts of research misconduct that goes beyond plagiarism. Even when elements of fabrication could be reasonably ruled out, it might still count as falsification.  

Cognitive Offloading and Not Cognitive Extension

When one uses a short prompt to instruct AI to generate a manuscript, it represents a shortcut taken to amass a large amount of knowledge from diverse sources and arranging this into a comprehensible and presentable form suitable for peer review/publication. A manuscript generated in such a manner entails obvious human-to-machine research task offloading, and the product might thus, in a perhaps overcritical sense, be considered as nonauthentic, or fabricated. One could counter this somewhat extreme notion by arguing that AI assistance is simply a form of cognitive or mind extension.14 As such, one’s use of AI in manuscript writing is analogous to the use of an electronic calculator in performing arithmetical tasks of considerable time and effort-consuming complexity.15 

However, there is a difference between cognitive extension with AI and cognitive offloading to AI. The former utilizes AI to expand one’s intellectual capacity in collecting, consolidating, and analyzing data/knowledge, while the latter simply delegates (at the expense of one’s intellectual effort) these tasks to AI. While the former could be considered as acquisition of knowledge ownership, the latter cannot, despite the human ultimately assuming authorship of the manuscript and not the AI (the latter is, indeed, widely prohibited). This is one good reason for arguing that the human author might have plagiarized from AI. Furthermore, to have acquired knowledge, the human author must make further intellectual investments to, at the very least, verify that the AI-generated draft is accurate. A failure to verify accuracy, coupled with improper declaration of AI use would, as further argued below, constitute falsification of research.  

How Much AI Use, and How Such Use Should Be Declared?

There is considerable variation among publishers and journals as to whether and how much AI use in assisting academic writing should be declared, and how such declarations should be worded. My position is that any and all use of AI should be declared, and declarations should be precise to the point of allowing the editors, reviewers, and readers to distinguish between the author’s original ideas, data/results, or factual content and manuscript draft, before and after modifications by AI. I would broadly divide such declarations into two levels, I and II.

The first level (I) would be what Resnik and Hosseini designated as substantial use—when AI is used to make decisions that directly affect research results and to generate and/or analyze content, data, or images.6 Such use should not only be declared but also documented in detail as integral parts within the writings in the relevant sections of the manuscript. For example, if AI is used for generation of ideas (such as identification of knowledge gaps), these must be detailed explicitly in the Introduction and Results (and/or Discussion) sections. If AI is used for the analysis of data, this should be described under Methods and/or Discussion. The above would be analogously equivalent to the descriptions necessary for the reader to understand how specific material composites were fabricated to exhibit superconductivity or how focus groups are organized and conducted to address a particular social issue. In other words, such AI use cannot be merely declared but should be explicated in a precise manner that allows full understanding of the science concerned. Any attempt to cover or obfuscate such details would misrepresent the research.

The second level (II) might be considerably more common, namely, the use of AI to assist or facilitate manuscript writing. Typically, these would include bibliometric collections/citations, language polishing, and presentation enhancements (including the drawing of schematic diagrams, etc). For these, many have been getting away with 1-sentence declarations with no details, as details of this type of AI use are often not mandated by journals and publishers. Considering that entire manuscripts of reviews, perspectives, opinions, and other types of articles that are not based on primary data could be readily generated by AI based on very simple prompts, it is clear that a brief declaration of AI use would be grossly inadequate. My position is that such declarations should at least be sufficiently detailed to accurately represent the research process and outcome. For example, if an author indicates that AI is used only for language polishing of a first draft, the prompts used, and the original human-written draft should be provided as supplementary materials for reference. If it was subsequently discovered that the manuscript was in fact drafted by AI in the first instance, or that AI had also generated some schematic diagrams, misrepresentation of research has occurred.  

Misrepresentation of Research—Falsification

A widely adopted definition of research misconduct is one formulated by the US Office of Research Integrity (ORI), and involves acts of fabrication, falsification, and plagiarism (FFP). “Falsification,” as defined by ORI, involves “…manipulating research materials, equipment, or processes, or changing or omitting data or results such that the research is not accurately represented in the research record.” At level I of the AI use declaration, in which the human author has no (or insignificant) intellectual input, the human author of an AI-generated manuscript would have misrepresented the research in terms of authenticity. By not verifying the accuracy of the manuscript content before submitting it for publication, the research would have been misrepresented in terms of veracity. 

A human author that has made no significant input toward an AI-generated manuscript has thus effectively manipulated a machine in the research process to yield quick deliverables that have no certainty of being fully accurate. Further, by not declaring that AI was used in generating such deliverables, the research data/knowledge is misrepresented in terms of its origin. The degree of aberration could be worse if data/knowledge in the draft are not properly referenced or are erroneous. That generative AI tends to hallucinate and produce gross inaccuracies, or nonsensical content is well known, as typified by the exaggeration seen in the rat figure alluded to above.4 In the above situation, it is inescapable to surmise that falsification of research has occurred. Even if the human author did not intend for errors and inaccuracies to appear in the manuscript, their lack of intellectual effort and transparency had effectively misrepresented or falsified the research and reporting process.

What about the presumably more common level II of AI use declaration, in which AI is acknowledged to have assisted with, or otherwise facilitated, manuscript text writing or figure preparations? As emphasized above, such declarations also need to be detailed enough to accurately represent the research process and outcome. Inadequate or imprecise declarations, in which the prompt input and the AI output are obfuscated such that they could not be readily verified by editors, reviewers, and readers, are also misrepresentations of research. Some might view the activities of writing and illustration requiring level II of AI use declaration as merely auxiliary or subsidiary to the main research process. However, given that the research manuscript or paper is the most audience-fronting component of research, its comprehensibility, veracity, and transparency are paramount. 

Research Misconduct and Human Authorship Justification

In summary, undeclared or underdeclared use of AI, be it at level I or II, could potentially misrepresent research in terms of data/knowledge authenticity, veracity, research process, and intellectual credit allocation. Such a machine-aided shortcut into producing unreliable research for publication would be a case of research misconduct. In accordance with this view, all academic journals should not only mandate declarations in AI use for manuscript preparation, but for such declarations to provide sufficient details to allow editorial assessment of authenticity, veracity, and credit on the part of the human author. In other words, the intellectual contribution on the part of the human author must be transparent, which would be in accordance with the ICMJE guidelines for authorship.16

While reviewers and editors are likely to simply decline a manuscript in which undeclared AI use has been detected during the screening or peer review process, such rejections are more often viewed and phrased as noncompliance with journal policy rather than outright research misconduct violations. Furthermore, if the paper is published and undeclared AI used is subsequently alleged by postpublication peer reviews, the elicited editorial responses of paper withdrawal or retraction may be more subdued or reserved than when acts of FFP are implicated. For example, in the former situation, the journal may opt to not escalate the matter with the author’s institution, and in the latter case the journal may choose to allow the authors to make retrospective declarations in corrigenda rather than retraction. 

As such, to some at least, the notion of undeclared or underdeclared levels I and II AI use being a form of research falsification may sound extreme or even far-fetched. However, if the intellectual contribution of human authors is so readily obscured in a manuscript with AI-generated components, academic authorship and even scholarship as we know it would be abysmally undermined. My view is that such a notion (if incorporated into journal policy) would serve a strong deterrence function, particularly against authors with implausibly high number and frequency of manuscript submissions/publications, possibly through their engagement of AI-based papermills.17 In this regard, actions in line with a detection of research misconduct, such as informing the authors’ institutions of the violations and a perceived threat of retractions rather than corrections would likely reduce cases of AI-based mass publication fraud. It would also likely deter the explosion of likely AI-generated formulaic research articles with inappropriate study designs and false discoveries based on public databases.18

References and Links

  1. Strzelecki A. ‘As of my last knowledge update’: how is content generated by ChatGPT infiltrating scientific papers published in premier journals? Learn Publ. 2024;38(1):e1650. https://doi.org/10.1002/leap.1650. 
  2. Glynn A. Suspected undeclared use of artificial intelligence in the academic literature: an analysis of the Academ-AI Dataset. ArXiv 2024;2411.15218. https://doi.org/10.48550/arXiv.2411.15218. 
  3. Kwon D. Science sleuths flag hundreds of papers that use AI without disclosing it. Nature 2025;641:290–291. https://doi.org/10.1038/d41586-025-01180-2. 
  4. Kaplan MH. On rats with oversized genitalia and other submissions. Immunohorizons 2024;8(3):227. https://doi.org/10.4049/immunohorizons.2400020. 
  5. Staiman A. 2025. When declarations just don’t cut it: building a risk-based framework for AI guidelines in publishing. Sci Ed. 2025;48:10–11. https://doi.org/10.36591/SE-4801-05.  
  6. Resnik DB, Hosseini M. Disclosing artificial intelligence use in scientific research and publication: When should disclosure be mandatory, optional, or unnecessary? Account Res. 2025. https://doi.org/10.1080/08989621.2025.2481949. 
  7. Popkov AA, Barrett TS. AI vs academia: experimental study on AI text detectors’ accuracy in behavioral health academic writing. Account Res. 2024. https://doi.org/10.1080/08989621.2024.2331757. 
  8. Gao R, Yu D, Gao B, Hua H, Hui Z, Gao J, Yin C. Legal regulation of AI-assisted academic writing: challenges, frameworks, and pathways. Front Artif Intell. 2025;8:1546064. https://doi.org/10.3389/frai.2025.1546064. 
  9. https://ori.hhs.gov/definition-research-misconduct
  10. Kwon D. AI is complicating plagiarism. How should scientists respond? Nature. 2024. https://doi.org/10.1038/d41586-024-02371-z. 
  11. Tang BL. The underappreciated wrong of AIgiarism – bypass plagiarism that risks propagation of erroneous and bias content. EXCLI J. 2023;22:907-910. https://doi.org/10.17179/excli2023-6435.
  12. Shaw D. The digital erosion of intellectual integrity: why misuse of generative AI is worse than plagiarism. AI & Soc. 2025. https://doi.org/10.1007/s00146-025-02362-2. 
  13. Yin S, Huang S, Xue P, Xu Z, Lian Z, Ye C, Ma S, Liu M, Hu Y, Lu P, et al. Generative artificial intelligence (GAI) usage guidelines for scholarly publishing: a cross-sectional study of medical journals. BMC Med. 2025;23(1):77. https://doi.org/10.1186/s12916-025-03899-1.
  14. Clark A. Supersizing the mind: embodiment, action, and cognitive extension, philosophy of mind series (online). Oxford Academic; 2009. https://doi.org/10.1093/acprof:oso/9780195333213.001.0001. 
  15. Tang BL. Will widespread use of artificial intelligence tools in manuscript writing mark the end of human scholarship as we know it? Sci Ed. 2025;12:231–233. https://doi.org/10.6087/kcse.366   
  16. International Committee of Medical Journal Editors.  Defining the role of authors and contributors [accessed August 28, 2025]. https://www.icmje.org/recommendations/browse/roles-and-responsibilities/defining-the-role-of-authors-and-contributors.html 
  17. Liverpool L. AI intensifies fight against ‘paper mills’ that churn out fake research. Nature. 2023;618(7964):222–223. https://doi.org/10.1038/d41586-023-01780-w. 
  18. Suchak T, Aliu AE, Harrison C, Zwiggelaar R, Geifman N, Spick M. Explosion of formulaic research articles, including inappropriate study designs and false discoveries, based on the NHANES US national health database. PLoS Biol. 2025;23(5):e3003152. https://doi.org/10.1371/journal.pbio.3003152.

 

Bor Luen Tang (https://orcid.org/0000-0002-1925-636X) is with the Department of Biochemistry, Yong Loo Lin School of Medicine, National University Health System, National University of Singapore.

Opinions expressed are those of the authors and do not necessarily reflect the opinions or policies of their employers, the Council of Science Editors, or the Editorial Board of Science Editor.