Free scoping read. A written assessment of your manuscript in two business days, at no cost.

Start one
Contents

Academy · Research Publication · AI's Impact on Academic Publishing

Challenges in Research and Publication: The Impact of AI on Research Integrity in Academia

More than 10,000 research papers were retracted in 2023, a new annual record according to Nature’s own analysis. Generative AI misuse, ranging from fabricated references to undisclosed drafting, is now a documented contributor to a retraction crisis that predates ChatGPT but has accelerated alongside it.

Pubrica AcademyWritten by a research integrity analyst Reviewed bya scholarly publishing reviewer Published10 October 2026 Last checked10 October 2026 Reading time8 minutes Sources10 peer-reviewed

The arrival of generative AI in academic research and publication has created challenges that extend well beyond the narrow question of whether an AI tool can be listed as an author. It has intersected with an already worsening retraction landscape, created new categories of undisclosed-use misconduct that existing editorial safeguards were not designed to detect, and prompted journals, institutions and funders into a rapid and still incomplete policy response. These challenges are best understood as a connected set rather than as isolated incidents if the effect of AI on the integrity of the published scientific record is to be assessed accurately.

Learning objectives

  • How the retraction landscape was already changing before generative AI became widely available
  • The specific, documented failure modes that generative AI has introduced into manuscript preparation
  • The extent to which editorial and institutional policy has kept pace, and the limits of that response

Common misconceptions

  • “The recent surge in research paper retractions is mainly a generative-AI problem.” Nature's own analysis found that more than 10,000 research papers were retracted in 2023, a new annual record; however, this increase was already underway before generative AI tools became widely available and was driven substantially by paper mills and systematic fraud in specific high-volume journals. Generative AI has added new, documented failure modes to this pre-existing trend rather than being its sole or original cause.

The retraction landscape before generative AI

Richard Van Noorden's analysis for Nature, drawing on Retraction Watch data, found that more than 10,000 research papers were retracted in 2023, a new annual record, with the increase driven substantially by a small number of high-volume journals and by the activity of paper mills (organizations that mass-produce fabricated or low-quality manuscripts for sale to researchers under pressure to publish)1. This increase in retractions was already underway and accelerating before generative AI tools such as ChatGPT became widely available in late 2022. This documented context is essential to scoping the contribution of AI correctly: the underlying pressures driving mass retraction (publication-volume incentives, paper-mill activity and uneven editorial screening at a subset of journals) predate generative AI and would persist in its absence1. Generative AI has intersected with this existing landscape, making certain kinds of fabrication faster and, in some documented cases, more difficult to detect with previously reliable screening methods1.

Documented failure modes introduced by AI

Within this landscape, generative AI has introduced specific, independently documented failure modes rather than a generic increase in misconduct. A controlled study by Gao and colleagues found that ChatGPT-generated scientific abstracts were misidentified as human-written by blinded human reviewers 32% of the time and were not flagged by standard plagiarism-detection software, demonstrating a concrete detection gap in a safeguard (plagiarism checking) that journals had in place before generative AI existed2. Separately, Stokel-Walker documented published instances of ChatGPT being listed as a credited co-author, a failure mode that required editorial bodies to write new policy because no existing authorship guideline had anticipated a non-human entity being proposed as an author3. A further, distinct failure mode documented in subsequent research and reporting is the generation of fabricated or inaccurate citations: a language model asked to produce references may cite papers that do not exist or misstate the content of papers that do. This failure mode has no equivalent in traditional plagiarism or image-manipulation misconduct, and it requires a form of verification that editorial screening processes were not designed to perform at scale, namely confirming that every cited source exists and supports the statement for which it is cited4.

Editorial policy responses and their limitations

By the standards of academic governance, the editorial and institutional policy response has been rapid: ICMJE, COPE and WAME each issued specific AI-related guidance within months of the release of ChatGPT, converging on the position that AI tools cannot be authors but that their use must be disclosed. Major publishers, including Elsevier and Springer Nature, subsequently issued operational policies requiring specific disclosure statements567. The limitations of this response are structural rather than a lack of effort: disclosure policies rely on authors to report AI use voluntarily, and the detection gap that Gao and colleagues measured in human review of AI-generated text indicates that an author who does not disclose AI assistance has a measurable chance of passing undetected through existing review processes2. Purpose-built AI-detection tools exist and are improving, but no detection tool has been shown in controlled, peer-reviewed study to identify AI-generated or AI-assisted text with sufficient reliability for a journal to treat automated detection as an adequate safeguard on its own. For this reason, current policy continues to rely heavily on author disclosure and reviewer judgment rather than on technical detection alone.

Implications for researchers

For researchers, the principal implication of these connected challenges is that the burden of verification has shifted onto the author. Whereas a plagiarism checker and an attentive reviewer were previously adequate safeguards against most forms of textual misconduct, the documented AI-specific failure modes (fabricated citations, fabricated abstracts passing human review a third of the time, and the impracticability of relying on detection tools alone) mean that an author using AI assistance must independently verify every factual claim, the existence and content of every citation, and every number the tool produced. AI-assisted drafts cannot be treated as exempt from the verification process that a fully human-written draft would require24. For the same reason, current editorial guidance consistently places final accountability for accuracy on the human authors, regardless of which tool assisted with which part of the manuscript: the verification burden has not moved to the AI tool or to the journal's screening process but has been placed explicitly on the author56.

Institutional and funder responses

Journal-level editorial policy is only one layer of the response to these challenges; research institutions and funding bodies have separately begun to address the role of generative AI in research conduct and reporting, sometimes with requirements that exceed those of individual journals. A growing number of universities have issued institutional guidance on acceptable AI use in student and faculty research output. This guidance is generally built on the same core distinction that journals have adopted (disclosure is required, and accountability for accuracy rests with the human researcher) but is applied earlier in the research process, at the stages of grant writing, data analysis and internal review, rather than only at journal submission. Research integrity offices at these institutions have correspondingly had to update their misconduct investigation procedures to address AI-specific scenarios, such as whether an undisclosed AI-fabricated citation in a grant application or thesis constitutes the same category of misconduct as traditional fabrication or falsification, a question that did not require a specific answer before generative AI tools became widely available.

Funders have moved in a similar direction: a number of major research funding bodies have added or updated provisions in their grant terms and conditions addressing AI tool use in proposal preparation and in the research itself. These provisions generally require the same disclosure principle that journals have adopted and reserve the right to treat undisclosed AI-generated fabrication in a funded project as a research-integrity violation subject to the funder's existing misconduct procedures. Because this institutional and funder-level policy layer is newer and less standardized than the now-converged journal-editorial position, researchers working across multiple institutions or funding sources, as is common in collaborative or multi-site research, should consult each relevant institution's and funder's current policy separately rather than assuming that compliance with a target journal's AI disclosure policy satisfies every other organization's requirements. For multi-site collaborations, the most reliable approach is to apply the strictest of the applicable institutional, funder and journal policies across the whole project from the outset, rather than tracking which contributor is bound by which policy for which section of a jointly authored manuscript; the latter approach produces a fragmented compliance record that is difficult to defend if any part of it is later challenged. Collaborators should also agree on this approach explicitly and in writing before drafting begins, since resolving a policy conflict after a manuscript has been written is considerably more disruptive and time-consuming than settling it as a condition of the collaboration. Publisher-level policy offers an instructive contrast: Holden Thorp's January 2023 editorial committing Science to a specific, publicly stated position, and the detailed operational policies that Springer Nature and Elsevier have since published, are each single, centrally maintained documents that a researcher can consult directly, a degree of clarity that the more fragmented institutional and funder layer has not yet consistently achieved8910.

Key points

Key points

Generative AI has intersected with, and in specific documented ways accelerated, an already worsening retraction landscape. AI-assisted drafts therefore require the same, and in some respects greater, independent verification of every citation, claim and number as a manuscript written entirely by humans.

Practical application

Practical application

  1. Every citation generated by an AI tool should be independently verified before submission. Authors should confirm that each cited source exists and accurately reflects what it is cited as saying; fabricated or misattributed citations are a documented failure mode specific to AI.
  2. Plagiarism checking or cursory human review should not be relied upon to detect AI-generated inaccuracies. Controlled studies have measured substantial detection gaps in both; disclosure and independent fact-checking remain necessary regardless of the review safeguards already in place.
  3. AI-assisted drafting should be treated as requiring more verification time, not less. Current editorial policy places responsibility for accuracy explicitly on the human author, so the same or greater fact-checking time should be allocated to AI-assisted sections as to sections drafted entirely by humans.

Frequently asked questions

Not primarily. Nature's own analysis found that the increase (more than 10,000 retractions in 2023, a new record) was already underway before generative AI tools became widely available and was driven substantially by paper mills and systematic fraud at specific high-volume journals, although generative AI has since added new, documented failure modes to this existing trend.

The term refers to an AI tool generating a reference to a source that does not exist, or misstating the content of a source that does exist. It is a documented failure mode distinct from traditional plagiarism, since no existing text is copied.

No tool has been shown in controlled, peer-reviewed study to do so with sufficient reliability for a journal to treat automated detection as adequate on its own, which is why current policy relies primarily on author disclosure rather than on detection technology.

Current ICMJE, COPE, and WAME guidance places final responsibility for accuracy on the human authors regardless of which tool contributed to drafting any specific part of the manuscript.

No. The policy response has been rapid in establishing disclosure requirements, but the underlying detection gap and the reliance on voluntary author disclosure remain structural limitations that current policy has not fully resolved.

References

  1. Van Noorden R. More than 10,000 research papers were retracted in 2023 — a new record. Nature. 2023;624:479-481. read 10 October 2026. View on Nature News
  2. Gao CA, Howard FM, Markov NS, Dyer EC, Ramesh S, Luo Y, Pearson AT. Comparing scientific abstracts generated by ChatGPT to original abstracts using an artificial intelligence output detector, plagiarism detector, and blinded human reviewers. npj Digital Medicine. 2023;6:75. read 10 October 2026. View on npj Digital Medicine
  3. Stokel-Walker C. ChatGPT listed as author on research papers: many scientists disapprove. Nature News. 2023;613:620-621. read 10 October 2026. View on Nature News
  4. Else H. Abstracts written by ChatGPT fool scientists. Nature News. 2023;613:423. read 10 October 2026. View on Nature News
  5. International Committee of Medical Journal Editors. Recommendations for the Conduct, Reporting, Editing, and Publication of Scholarly Work in Medical Journals: Defining the Role of Authors and Contributors. read 10 October 2026. View on ICMJE.org
  6. Committee on Publication Ethics (COPE). Authorship and AI Tools: COPE Position Statement. 2023. read 10 October 2026. View on COPE
  7. World Association of Medical Editors. WAME Recommendations on Chatbots and Generative Artificial Intelligence in Relation to Scholarly Publications. 2023. read 10 October 2026. View on WAME
  8. Thorp HH. ChatGPT is fun, but not an author. Science. 2023;379(6630):313. read 10 October 2026. View on Science
  9. Springer Nature. Artificial Intelligence (AI) Policy for Authors. read 10 October 2026. View on SpringerNature.com
  10. Elsevier. The Use of AI and AI-assisted Technologies in Writing for Elsevier. read 10 October 2026. View on Elsevier.com

Every source above is a peer-reviewed article, a publisher’s record of one, or a standard issued by the body that owns it. Each carries the date it was last read. Where a standard moves, this article is revised rather than left up with a note.

How to cite this article

Pick your referencing style and copy it. Your university or target journal may vary the punctuation — check its guide if it has one.

Pubrica Academy. (2026, October 10). Challenges in Research and Publication: The Impact of AI on Research Integrity in Academia. Pubrica. https://pubrica.com/academy/citation-and-formatting/challenges-of-ai-in-research-and-publication/

This article is written to be learned from. Cite it if you are referring to what it says; if you are using the evidence behind it, cite the sources listed above instead.

AI-Assisted Manuscript Verification Checklist

Run this against any manuscript section drafted with AI assistance before submission.

[ ] Independently verified every citation generated or suggested by an AI tool actually exists
[ ] Confirmed each verified citation's content matches what it is cited as supporting
[ ] Fact-checked every number, statistic, or claim the AI tool produced against the source data
[ ] Disclosed AI use per the target journal's specific current policy
[ ] Did not rely on plagiarism-checking software as a substitute for AI-content verification
[ ] Confirmed no AI tool is credited as an author or contributor

Pubrica AcademyWritten and maintained by Pubrica’s subject matter experts. Articles are revised when the underlying standards change, and each article states the date on which it was last reviewed.

How Pubrica can help

Pubrica’s research integrity team independently verifies AI-assisted manuscript content (citations, claims and figures) against source material before submission, addressing the specific detection and verification gaps documented in current research on generative AI in academic publishing.

Find something else

Search the Pubrica Academy

Access is free and requires no account. Enter a topic, or open the full library to filter by category, sub-topic or resource type.

Verification of AI-assisted claims and citations in a manuscript

A scoping assessment is free and takes two business days. It provides a written assessment of what the work requires and which specialist would undertake it.

Get a free quote

Regulations

Check your project against this rule

  1. 1Your work
  2. 2Your details
  3. 3Send
What do you need?

Material you send is held under signed confidentiality, with access limited to the assigned team and encrypted transfer. It is not used to train anything and is not passed to third-party tools. Ask for the agreement before you send anything you would rather cover first.