Can You Upload a Collaborator’s Unpublished Work to AI Without Permission?

If a Collaborator Shared It With You, Can You Share It With AI?

A collaborator emails you a draft manuscript and asks for feedback. A co-investigator shares preliminary results in a project folder. A doctoral student sends you an unpublished chapter. A colleague gives you code for a joint analysis.

You legitimately possess the material. An AI tool could help summarize it, debug it, improve the writing, or identify weaknesses. Can you upload it without asking the person who shared it?

Not necessarily. Permission to access unpublished work for one purpose is not automatically permission to disclose it to an external AI service for another.

Access Does Not Automatically Mean Permission to Upload

The Missing Question Is Often Authority, Not Technology

Sharing something with you does not make it public

Researchers routinely exchange unpublished work with collaborators. That sharing usually occurs within a relationship and for a purpose: review this draft, run this analysis, check these calculations, contribute to this paper, or prepare this presentation.

The material does not thereby become public or available for unrestricted onward disclosure.

Submitting it to an externally operated AI service can introduce another party or processing environment into the workflow. Whether that is acceptable depends on what the collaborator authorized, explicitly or through the project's agreed arrangements.

Possession, authorship, and authority are different concepts

You may have a copy of a file without owning it. You may be a co-author without controlling every contribution inside it. You may own some intellectual property while the university, funder, sponsor, or another collaborator has rights in other parts.

The first does not automatically establish the second.

Collaborative research often contains overlapping rights and obligations

A single manuscript or dataset can combine contributions from many sources.

This is why a blanket statement such as "I am a co-author, so I can upload it" is unreliable.

Permission can depend on what you want AI to do

Not every AI use exposes the same amount of material. Asking an AI assistant to improve one generic sentence differs from uploading a collaborator's entire unpublished dataset and asking for exploratory analysis.

A collaborator might be comfortable with language editing in an institutionally approved AI environment but object to their raw data being uploaded to a public chatbot. Permission should therefore be sufficiently specific to the proposed use when the distinction matters.

Confidentiality can exist without a formal NDA

Researchers sometimes assume that information is confidential only when a document has "CONFIDENTIAL" stamped across the first page or a formal nondisclosure agreement exists.

Collaborative research is often shared under professional and project-specific expectations that unpublished drafts, preliminary findings, research ideas, and internal discussions will remain within the team until the group decides otherwise. Formal agreements can strengthen or define those obligations, but the absence of an NDA does not automatically make unpublished material suitable for external disclosure.

At minimum, uncertainty about a collaborator's expectations is a reason to ask rather than a reason to assume permission.

Removing the collaborator's name does not necessarily solve the problem

If the concern were solely personal identification, removing names might reduce some risk. But unpublished collaborative work can be sensitive for reasons unrelated to identity.

The material may contain a novel method, preliminary finding, research idea, proprietary dataset, patentable invention, unpublished code, or confidential partner information. All of those can remain sensitive after the author's name disappears.

Even where identification is the concern, contextual information may still reveal the researcher, institution, project, or participants. The limitations of removing names before submitting research information to AI therefore apply here too.

A no-training setting does not create permission

Suppose the AI provider promises not to use your prompts to train its models. That may make the service more acceptable for some research workflows, but it cannot grant rights over somebody else's unpublished work.

If your collaborator has not authorized external processing, changing the AI account's training settings does not supply the missing permission. Nor does it resolve any institutional, contractual, participant, or intellectual-property restrictions attached to the material.

Participant data can make collaborator permission insufficient

There is another layer. Even if your collaborator says, "Sure, upload it," that permission may not be enough when the material contains research-participant information or data governed by ethics approval, consent, data-use agreements, privacy law, or institutional security requirements.

A collaborator cannot necessarily authorize a use that the research protocol itself does not permit.

The decision must therefore account for both interpersonal permission and the formal governance of the research information.

Institutional and sponsor agreements may control the decision

Collaborations involving multiple universities, companies, hospitals, government agencies, or external funders often operate under agreements defining confidentiality, publication, data use, intellectual property, and authorized systems.

Where such agreements exist, informal approval from one collaborator may not override them. Researchers may need to consult the principal investigator, data custodian, research office, technology-transfer office, legal office, information-security team, or another designated authority.

Peer review is an especially clear example of limited-purpose access

The logic becomes easiest to see in peer review. A journal sends a reviewer another researcher's unpublished manuscript. The reviewer has legitimate access to it, but that access exists solely for confidential evaluation. It does not create permission to upload the paper elsewhere.

Major publishing guidance accordingly restricts disclosure of peer-review material to AI. The same conceptual distinction is useful in collaboration: authorized access for a particular research purpose does not imply unlimited onward disclosure.

The stricter case is discussed separately in using confidential peer-review material with AI.

Sometimes you can accomplish the task without the collaborator's material

Before asking for permission to upload a 50-page draft, ask whether the AI actually needs it.

If you need advice about structuring a discussion section, you may be able to describe the structural problem abstractly. If you are debugging code, a minimal reproducible example may remove the proprietary project logic. If you need methodological guidance, a synthetic dataset can sometimes reproduce the analytical problem without exposing the original data.

Minimization does not replace permission when the remaining material still belongs to the collaborator, but it can sometimes eliminate the need to disclose their work at all.

Make AI expectations part of collaboration agreements early

Repeatedly asking permission for every minor AI-assisted task can become cumbersome in a long-term project. Research teams can reduce ambiguity by discussing AI use when establishing their workflow.

A team might agree which AI systems are approved, what categories of information may be submitted, whether full manuscripts can be uploaded, whether participant data are prohibited, what uses require consultation, and how AI assistance will be documented or disclosed.

That conversation turns an implicit assumption into an explicit research practice. As a bonus, it may prevent the traditional academic discovery that five co-authors had six different understandings of the workflow.

A Collaborator Sends You a Draft for Feedback

The important step was recognizing that "please review my draft" did not silently contain the additional instruction "and feel free to disclose it to whichever AI service you prefer."

Common Mistakes About Collaborator Permission and AI

Ask Who Authorized the Disclosure Before Asking What AI Can Do

The central issue is not whether your collaborator trusts you. It is whether the proposed AI disclosure falls within the permissions and safeguards governing the shared material.

This approach is especially useful when working with co-authored unpublished manuscripts, where a single file can embody the intellectual contributions and obligations of an entire research team.

Before Uploading a Collaborator's Unpublished Work to AI

Questions About Collaborator Permission and AI

Being Trusted With Unpublished Work Does Not Make It Yours to Redistribute

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.