03 · What You Need to Know
Why Uploading a Peer-Review Manuscript to AI Requires Special Permission
Why Is a Manuscript Confidential During Peer Review?
A manuscript submitted for journal consideration is ordinarily treated as a privileged communication within the editorial process. Reviewers receive access because they have been invited to provide an expert assessment, not because the manuscript has become publicly available.
The document may contain unpublished findings, original research ideas, confidential datasets, preliminary intellectual property, or information that could identify research participants.
Confidentiality protects the authors' interests and supports trust in the editorial process. It also limits how reviewers may use or disclose information obtained through their assignment.
The International Committee of Medical Journal Editors (ICMJE) emphasizes that submitted manuscripts are privileged communications and that reviewers must maintain their confidentiality.
These obligations apply even when the reviewer believes that sharing the manuscript with a technological service would improve the quality or efficiency of the assessment.
Why Is Uploading a Manuscript to AI Different From Reading It Yourself?
When a reviewer reads a manuscript within the journal's authorized review process, the document remains within the access arrangements established for that assignment.
Uploading it to an external AI service may introduce another organization and its technical infrastructure into the handling of the manuscript.
Depending on the service, the document may be transmitted to remote servers, processed through additional systems, retained for operational purposes, or made accessible to authorized service personnel under particular circumstances.
These possibilities depend on the provider's actual contractual terms and technical configuration. Researchers should not assume that every AI service handles uploaded material identically.
The ethical concern is that the reviewer may be extending access to unpublished research beyond what the journal and authors authorized.
What Do Major Publishing Authorities Say?
There is no single worldwide rule governing every journal, but several influential publishing organizations have established restrictions or expectations concerning AI use during peer review.
| Authority |
Relevant Position |
Practical Implication |
| ICMJE |
Reviewers should follow journal AI policies or request permission and must protect manuscript confidentiality. |
Do not assume AI use or manuscript uploading is authorized. |
| Elsevier |
Its published reviewer policy restricts uploading manuscripts or portions of manuscripts to generative AI tools because of confidentiality and proprietary-rights concerns. |
Follow the current publisher and journal instructions before considering any AI processing. |
| Springer Nature |
Its AI editorial guidance emphasizes confidentiality and restricts uploading confidential manuscripts into generative AI tools. |
Do not transfer submitted manuscript material to unauthorized services. |
| COPE |
Its ethical guidance treats confidentiality and responsible handling of unpublished submissions as core peer-review obligations. |
Technological assistance does not remove the reviewer's duty to protect the submission. |
These descriptions summarize published policy positions and ethical principles. The precise wording, exceptions, and permitted tools may change. Reviewers should consult the current policy governing their specific assignment rather than rely on a general summary.
The ICMJE recommendations are influential, particularly in biomedical publishing, but they do not automatically govern every journal. Publisher-wide policies and journal-specific instructions may also differ.
Does a No-Training Promise Make Uploading Safe?
Not necessarily.
Whether uploaded information is used to train future models is only one aspect of data handling.
Other relevant questions concern whether the provider retains the material, how long it remains accessible, who can access it, where it is processed, whether subprocessors are involved, and what contractual protections apply.
A service may exclude customer data from model training while still processing or retaining it for other authorized operational purposes.
More importantly, even a technically secure service may be prohibited by the journal's peer-review policy.
Reviewers should distinguish between a provider's data-handling assurances and the journal's authorization to disclose confidential manuscript material.
Provider data protection
Concerns how the AI service processes, stores, protects, and potentially uses submitted information.
Journal authorization
Determines whether the reviewer is permitted to disclose manuscript material to the particular system for the intended purpose.
Both may be necessary. Neither automatically substitutes for the other.
Does Removing the Authors' Names Make the Manuscript Safe to Upload?
No. Removing author names does not necessarily remove confidential or identifying information.
A manuscript may contain distinctive research questions, institutional settings, specialized datasets, unpublished results, or references to ongoing projects that make the work identifiable.
Even if the authors cannot be identified, the unpublished scientific content may remain confidential.
Anonymization and confidentiality are therefore different issues. A document can contain no direct personal identifiers while still being subject to editorial restrictions.
Reviewers should not assume that a blinded manuscript is freely transferable to an external service.
What About Uploading Only One Paragraph, Table, or Figure?
Confidentiality obligations generally concern the manuscript's information, not merely the complete document as a file.
A single table may disclose unpublished findings. A methods paragraph may reveal a novel procedure. A figure may contain confidential participant information or commercially sensitive results.
Consequently, uploading an excerpt can raise the same underlying concern as uploading the entire manuscript.
Even paraphrasing a distinctive unpublished finding into an AI prompt may disclose protected information.
The relevant question is whether the material shared is confidential and whether the journal has authorized its processing.
Are Institutional or Enterprise AI Tools Different?
Institutionally managed AI services may provide stronger contractual controls, restricted access, defined retention arrangements, or local processing capabilities.
These features can reduce particular data-handling risks, but they do not automatically authorize use of confidential manuscripts.
A university's permission to use an AI tool for institutional research does not necessarily override a journal's separate restrictions on peer-review material.
Similarly, a locally deployed model may avoid transmission to an external provider while still being subject to journal policies governing AI-assisted review.
Reviewers should verify both the applicable editorial rules and the actual technical safeguards before processing manuscript content.
Can the Editor Give Permission to Use AI?
Permission may be possible when the journal's policy allows it and the editor has authority to approve the proposed arrangement.
However, a general statement that AI assistance is acceptable should not automatically be interpreted as authorization to upload the entire manuscript to any service.
The reviewer should clarify the intended tool, the information to be processed, the purpose of processing, and any required confidentiality safeguards.
Where the journal has established an approved AI environment, reviewers should follow its specific instructions.
If authorization remains unclear, the safer course is not to upload the manuscript.
What If the Manuscript Is Already Available as a Preprint?
A publicly available preprint changes the status of the publicly released version, but it does not automatically remove confidentiality obligations concerning the journal submission.
The submitted manuscript may contain revisions, unpublished supplementary material, confidential reviewer communications, or other information not included in the preprint.
Moreover, a journal may restrict AI use during peer review regardless of whether similar material is publicly accessible.
Reviewers should distinguish the publicly available document from the confidential submission and follow the journal's policy.
Can AI Be Used Without Disclosing Manuscript Content?
Some general assistance may be possible without transmitting confidential information, provided the journal permits the intended AI use.
For example, a reviewer might seek a generic explanation of a statistical method or request a general checklist for evaluating an observational study without revealing anything about the assigned manuscript.
However, the reviewer must avoid including distinctive manuscript details in the request.
Whether generative AI may assist with preparing the peer-review report is a separate question from whether the manuscript itself may be uploaded.
Permission for one activity does not necessarily authorize the other.
What If a Reviewer Has Already Uploaded the Manuscript?
If a confidential manuscript has been uploaded without authorization, the reviewer should not assume that deleting the conversation resolves the matter.
Depending on the provider, deletion may not immediately remove all retained copies or processing records.
The reviewer should stop further disclosure, preserve the information needed to describe what occurred, and promptly contact the handling editor or the journal's designated confidentiality contact.
Relevant institutional privacy or information-security personnel may also need to be involved, depending on the material and applicable requirements.
The response should follow the journal's incident procedures and the provider's documented data-handling arrangements. Reviewers should avoid making unsupported assurances about deletion or the absence of exposure.
