The rules are becoming increasingly clear. If you are invited to review a manuscript, you should not upload it to ChatGPT or any other generative AI system. You should not paste the text into an AI tool and ask for a summary, upload the results section and ask whether the analysis is convincing, or ask the system to identify weaknesses, generate reviewer comments or write the review for you. Editors are often subject to similar restrictions.
The reasons are understandable. A submitted manuscript is a confidential document. It may contain unpublished findings, commercially sensitive information, personal data or ideas that the authors have not yet made public. Uploading it to an external platform may expose the manuscript to another organisation and to systems whose data-retention practices are not always clear.
Publishers also raise concerns about copyright and intellectual property. Elsevier tells reviewers and editors not to upload a submitted manuscript, or any part of it, into an AI tool because doing so may violate the authors’ confidentiality and proprietary rights. Where personally identifiable information is included, it may also breach data-privacy rights.
Sage warns that the use of generative AI during peer review can breach peer-review confidentiality. Some Sage-published journals state the concern even more directly, describing the uploading of unpublished work as creating both confidentiality risks and copyright issues.
Wiley similarly prohibits editors from uploading manuscripts under review to AI detection tools, explaining that this may undermine both review confidentiality and author copyright. FEBS Press, published through Wiley, says that AI tools may use uploaded material for training or other purposes, potentially violating confidentiality, privacy and the copyright of the manuscript.
These policies are not unreasonable. Reviewers have been entrusted with somebody else’s work for a limited purpose. They have not been given unrestricted permission to share it with other people, organisations or technological systems. However, there is an awkward question that publishers do not always appear to ask.
Has the author already uploaded it?
The manuscript that the reviewer is forbidden from uploading may already have been uploaded by the authors. Perhaps they asked an AI system to correct spelling and grammatical errors, improve the clarity of the argument, shorten the abstract or rewrite an awkward paragraph.
They might also have uploaded the whole paper and asked for suggestions. The AI system may have been asked to identify gaps in the literature, strengthen the discussion, suggest alternative titles, generate keywords or anticipate criticisms from reviewers. It may even have been asked to act as a peer reviewer before the paper was submitted to a journal.
Some of these activities may be permitted by the journal. Others may require disclosure, while some may be discouraged depending on the publisher, discipline or extent of the intervention. The important point is that they may already have happened.
This creates an unusual asymmetry. Once the paper enters the journal’s editorial system, the manuscript is treated as too confidential and too legally sensitive to be uploaded to an AI platform. Before submission, however, the authors may have uploaded the same paper, perhaps several times and in several different versions. The journal may never ask. Personally speaking, I have never been asked.
The difference is not trivial
There is an obvious response to this argument. The authors wrote the manuscript, or at least claim ownership of its content, and can therefore decide whether to upload it to an external system. The reviewer cannot make that decision because the manuscript does not belong to them.
That distinction matters. A reviewer receives a paper under conditions of confidentiality. The manuscript has been supplied for the specific purpose of evaluation. Passing it to an AI tool may amount to sharing it with a third party, even where the reviewer regards the system as no more than a writing or analytical tool.
Author consent would therefore address an important part of the problem, but it would not resolve everything. A paper may contain personal data, information supplied under an agreement with an external organisation, confidential material owned by another party or contributions from several co-authors who do not all share the same view.
There are also important differences among AI systems. Some public tools may retain prompts or use data to improve their services. Some enterprise systems offer stronger privacy safeguards, while some journals may eventually operate their own protected AI tools within controlled environments.
The choice is not simply between “AI” and “no AI.” It is also about which system is used, what information is entered, what happens to that information and who is responsible for the outcome. Nevertheless, journals could start by asking two questions.
1) Have the authors used AI?
This first question could be included during submission:
“Before submitting this manuscript, did the authors upload the manuscript, or a substantial part of it, to a generative AI system?”
This need not be a trap. Answering “yes” should not automatically lead to rejection. The purpose would be transparency, not punishment. Authors could be asked to explain briefly what tool was used and for what purpose.
A response might say that an AI tool was used to improve grammar and readability. Another might report that the manuscript was examined for structural weaknesses, while a third might disclose that AI was used to generate or rewrite substantial passages. Those are not necessarily equivalent uses, and journals may wish to treat them differently.
Many publishers already expect authors to disclose certain uses of generative AI. The difficulty is that the requirements vary, and authors may not always understand where routine language assistance ends and substantive intellectual contribution begins. A direct submission question would at least remove some of that ambiguity.
More importantly, it would tell the journal whether the manuscript has already been shared with an external AI system. That fact may be relevant when publishers justify their restrictions on editors and reviewers through confidentiality, copyright or data-security concerns.
2) Would the authors permit AI-assisted review?
This second question is potentially more controversial:
“Would the authors consent to the manuscript being processed by an approved generative AI system to support editorial assessment or peer review?”
The word “approved” is important. This should not give every reviewer permission to upload the paper to any public AI platform they happen to use. The journal would need to define which systems are acceptable, what safeguards are in place and what forms of assistance are permitted.
Authors might be offered several choices. They could decline all AI-assisted editorial or peer-review processing, permit it only through a secure system controlled or approved by the publisher, or agree to limited uses such as language analysis, reference checking or the identification of reporting omissions while refusing broader automated evaluation.
This would transform an assumed prohibition into an explicit choice. It would also force journals to explain what they mean by AI-assisted peer review. Are they considering tools that check statistical reporting, systems that identify missing references, software that compares a manuscript against reporting guidelines, or an LLM asked to provide a complete assessment of the paper? Those uses should not be treated as though they are identical.
Transparency Must Work Both Ways
Authors are increasingly expected to disclose their use of generative AI. That expectation is defensible because readers, editors and reviewers should know when a system has played a role in producing the manuscript. However, transparency should not stop when the paper is submitted.
Where an editor uses an AI system to summarise a manuscript, should the authors be told? If a reviewer uses an approved AI tool to help identify weaknesses, should that be disclosed? If a publisher runs an automated screening system over every submission, should that information appear in the journal’s policies?
If authors are expected to declare their use of AI, editors, reviewers and publishers should be expected to do the same. The journal might tell authors whether AI-supported tools were used during initial screening, editorial assessment or peer review. Reviewers could also be asked to declare any permitted AI assistance when submitting their reports.
This would not diminish human responsibility. The editor must remain responsible for the editorial decision, while the reviewer must remain responsible for the review. An AI system cannot be held accountable for misunderstanding a method, overlooking a conflict of interest, inventing a criticism or recommending rejection. Disclosure would simply make the process more visible.
Consent is necessary, but not sufficient
There is a danger that the proposal is reduced to a simple argument: if the authors agree, reviewers should be allowed to upload the manuscript. That would go too far.
Consent is one part of responsible governance. It does not remove the need for data protection, secure systems, clear contractual arrangements, human oversight and limits on the purposes for which a manuscript may be processed.
Nor should author consent become compulsory in practice. Authors may fear that refusing AI-assisted review will delay their submission or make editors less willing to handle it. Journals would need to ensure that declining consent does not create a hidden disadvantage.
Co-author agreement would also matter. The corresponding author should not casually give permission on behalf of collaborators without ensuring that they understand what is being proposed. There may also be manuscripts for which AI processing is inappropriate regardless of author preference, including papers containing identifiable patient information, sensitive interviews, commercially restricted data or material subject to legal or national-security controls.
The appropriate model is not unrestricted permission. It is informed consent within a governed system.
What is the real problem?
Publishers are right to protect unpublished manuscripts. Reviewers should not quietly transfer confidential papers to external platforms because doing so is convenient. However, the present approach risks becoming inconsistent.
A manuscript may be uploaded repeatedly by its authors during preparation, then treated as though it has never entered an AI system once it reaches the journal. Authors may be asked to disclose some uses of AI while receiving little information about the automated systems used by publishers. Reviewers may face an absolute ban even where the authors would be comfortable with limited processing in a secure environment.
The problem is not necessarily that an AI system has seen the manuscript. The problem is that nobody knows who uploaded it, what they uploaded, why they uploaded it, when they uploaded, what system they used or what happened to the information afterwards.
Transparency would help. Consent would help. Approved systems and clearly defined uses would help. Publishers should certainly tell reviewers what they must not do, but perhaps they should also ask authors what has already been done and what they would be prepared to permit.
That would move the discussion beyond prohibition and towards responsible governance. The question is not whether AI should be used in peer review. It is whether scholarly publishing can create a system in which authors, editors and reviewers know when it is being used, understand the risks and agree on the rules.
Publisher policies
Elsevier, Generative AI Policies for Journals:
https://www.elsevier.com/about/policies-and-standards/generative-ai-policies-for-journals
Sage, Artificial Intelligence Policy:
https://www.sagepub.com/journals/publication-ethics-policies/artificial-intelligence-policy
Wiley, Using AI tools in your research:
https://www.wiley.com/en-nl/publish/article/ai-guidelines/
Wiley, Editor FAQs: Wiley’s AI Guidelines for research authors, editors, and reviewers:
https://www.wiley.com/en-be/publish/editor-insights/editor-faqs-ai-researcher-guidelines/
FEBS Press, Peer Review Policies:
https://febs.onlinelibrary.wiley.com/hub/peer-review-policies
Did you find this edition useful?
I publish two free weekly newsletters exploring different aspects of higher education, research and scholarly publishing. If you enjoyed this edition, you may also find value in the following.
Governance & Research in HE
A weekly brief for higher education leaders and researchers examining governance, leadership and research in practice.
Publishing with Integrity
Examining scholarly publishing through an ethical lens, challenging assumptions and supporting academic career development.
About the Author
Professor Graham Kendall is Vice-Chancellor of GlobalNxt University, Malaysia, and an Emeritus Professor at the University of Nottingham. Over the past 25 years, he has published more than 300 peer-reviewed papers and has served as Editor-in-Chief and Associate Editor of several international journals. Through Publishing with Integrity, he explores the ethical, governance and practical challenges facing scholarly publishing, encouraging greater transparency, integrity and informed debate across the global research community.