Can a Grant Reviewer Upload a Confidential Research Proposal to an AI Tool?

Can You Upload Someone Else's Grant Proposal to AI for Review?

You have been invited to evaluate a research grant application. The proposal is lengthy, technically demanding, and contains detailed information about the applicant's research plans.

To save time, you consider uploading the document to a generative AI tool and asking it to summarize the methodology, identify weaknesses, or draft preliminary reviewer comments.

But the proposal is confidential. Does your appointment as a reviewer give you permission to share it with an AI system?

The issue is not limited to whether the AI provider might retain the document or use it for model training. Grant reviewers receive unpublished research information under specific obligations. Uploading that material to an external service may involve disclosure to a party that the applicant and funding organization never authorized.

Even seemingly limited activities, such as copying a proposal abstract or asking AI to evaluate an unpublished research idea, can raise confidentiality concerns.

Do Not Upload Confidential Proposals Without Explicit Authorization

Why Confidential Grant Proposals Require Special Protection

Why Are Research Grant Applications Confidential During Peer Review?

Grant applications frequently contain information that applicants have not publicly disclosed. This may include original research ideas, preliminary findings, methodological innovations, proposed collaborations, or plans for developing intellectual property.

Applicants submit this information so that a funding organization can evaluate the proposed research. Reviewers receive access for that specific purpose, usually under confidentiality agreements or equivalent obligations.

Those arrangements help protect applicants from unauthorized disclosure and support confidence in the fairness and integrity of the evaluation process.

The confidentiality obligation generally concerns more than the complete application document. It may also extend to information learned through the review process, reviewer discussions, critiques, scores, and other restricted records.

Consequently, the absence of an applicant's name or the removal of a proposal's title does not necessarily make the remaining material suitable for external disclosure.

Does Being Appointed as a Reviewer Give You Permission to Upload the Proposal?

No. Authorization to evaluate a proposal does not automatically include authorization to transmit it to third-party services.

A reviewer may be permitted to download an application through the funder's review system, read it, and prepare an assessment using approved procedures. Those permissions do not establish that the reviewer may upload the document to a commercial AI provider.

The relevant authority belongs to the funding organization and the applicable review arrangements.

Reviewers should therefore distinguish between authorized access to confidential material and authorized onward disclosure or processing.

What Does NIH Say About Uploading Grant Applications to AI?

In June 2023, the US National Institutes of Health (NIH) issued Notice NOT-OD-23-149, which prohibits generative AI use in the NIH scientific peer review process.

The notice specifically addresses the use of generative AI to analyze grant applications and formulate peer review critiques.

It also explains that uploading or sharing application content, original concepts, or review critiques with online generative AI tools violates NIH peer review confidentiality and integrity requirements.

These restrictions apply to the reviewer's handling of material received through the NIH peer review process.

The notice discusses accessibility-related circumstances requiring communication with the designated NIH official. Reviewers should follow the applicable authorization procedure rather than independently assuming an exception.

NIH's reviewer restriction should also be distinguished from its separate policies governing applicants' preparation of their own research proposals.

What Does NSF Say About Uploading Confidential Proposals?

The US National Science Foundation (NSF) issued a notice in December 2023 concerning generative AI in its merit review process.

NSF prohibits reviewers from uploading proposal content, review information, and related records to non-approved generative AI tools.

The distinction between approved and non-approved tools matters. The policy should not be misrepresented as a statement that every possible AI technology is prohibited in every NSF-related activity.

Nevertheless, a reviewer cannot independently declare a tool approved because it offers encryption, enterprise privacy controls, or a favorable data retention policy.

Any authorization must be established through NSF's applicable procedures.

Reviewers should consult the current notice and their review instructions before processing any confidential material through an AI system.

Which Information Should Reviewers Avoid Uploading?

Confidentiality risks are not limited to uploading a complete proposal file.

Whether a particular item is confidential depends on the applicable review agreement and the circumstances. However, reviewers should not assume that removing names, shortening passages, or paraphrasing technical content automatically makes disclosure permissible.

Why Is Uploading a Proposal Different From Reading It in an Approved System?

When reviewers access an application through an authorized review platform, the funding organization can establish conditions governing access, storage, permitted users, and handling procedures.

Uploading the same material to an external AI service may introduce another organization into the processing arrangement.

Depending on the service, submitted information may be stored, logged, processed by service providers, or retained under terms that differ from the funder's confidentiality requirements.

The critical question is whether that processing has been authorized.

Even where the AI provider offers substantial security protections, those protections do not automatically establish that the reviewer has permission to transmit the information.

Does a No-Training Policy Make Uploading Confidential Proposals Acceptable?

No. A commitment not to use submitted content for model training addresses one possible use of the information. It does not resolve every confidentiality question.

Researchers should distinguish among training use, operational retention, authorized service-provider access, data deletion, and contractual restrictions.

For example, an enterprise AI service may provide commitments concerning training and security while still processing information through infrastructure outside the funding organization's authorized review environment.

If the funder's rules prohibit the proposed activity or require specific authorization, the reviewer's choice of a no-training service does not override those rules.

Can Reviewers Use Temporary Chats or Delete the Conversation Afterward?

Temporary conversation features and deletion controls may reduce certain retention or visibility risks, depending on the service.

However, they do not establish that the original disclosure was authorized.

Deleting a conversation after uploading confidential material cannot retroactively provide permission for the transmission. Nor should reviewers assume that deleting a visible conversation immediately removes every associated operational record.

Reviewers must establish authorization before submitting confidential information, not attempt to resolve confidentiality concerns after the upload has occurred.

Can Reviewers Upload Only a Small Excerpt?

Not necessarily.

A short excerpt may contain the most sensitive information in an application. A few sentences describing an unpublished invention, experimental method, or research hypothesis could reveal information the applicant intended to keep confidential.

The amount of text is therefore not a reliable measure of disclosure risk.

Reviewers should assess whether the information originated from the confidential application and whether the funding organization's rules authorize its external processing.

Under policies that prohibit sharing application content with unauthorized AI systems, shortening the content does not establish an exception.

Can Reviewers Anonymize a Proposal Before Uploading It?

Removing the applicant's name, institution, or project title does not necessarily eliminate confidentiality concerns.

A research proposal may remain recognizable through its scientific approach, study population, collaborating institutions, preliminary results, or distinctive technical details.

More importantly, confidentiality can protect unpublished research information independently of whether the applicant is identifiable.

Anonymization may reduce certain personal data risks, but it does not automatically authorize the disclosure of confidential scientific content.

Can Reviewers Ask AI General Methodological Questions?

Potentially, depending on the funding organization's rules and whether the interaction reveals confidential information.

A reviewer might want to understand a statistical technique or clarify a general methodological concept. Consulting publicly available educational resources may be different from entering details obtained from a confidential proposal.

However, even a question framed without names can reveal restricted information if it reproduces distinctive features of the application.

Reviewers should avoid using confidential proposal details in prompts unless the funder explicitly authorizes that processing.

Where additional expertise is needed, the safer course is to use the funding organization's established procedures for consultation or specialist review.

What If the AI Tool Is Provided by the Reviewer's University?

Institutional approval of an AI service does not automatically establish permission to process another organization's confidential grant applications.

A university may authorize a tool for internal research administration or manuscript preparation while a funding agency separately prohibits its use for confidential peer review.

Both sets of obligations may be relevant, but the reviewer's access to the proposal is governed by the funder's review arrangements.

Reviewers should therefore obtain the required funder authorization rather than relying solely on their institution's approval of the software.

What Should a Reviewer Do After an Accidental Upload?

If confidential application material has already been uploaded to an unauthorized AI service, the reviewer should not assume that deleting the conversation resolves the issue.

The appropriate response is to follow the funding organization's incident reporting procedures and promptly notify the designated review official or security contact, as required.

The reviewer should preserve relevant factual information about what was disclosed, when it occurred, and which service was involved, while avoiding further unnecessary dissemination.

Any deletion, containment, or notification measures should follow the responsible organization's instructions.

The reviewer should not independently promise the applicant that no disclosure occurred or that the uploaded information has been completely removed unless those claims have been established.

When a Reviewer Considers Uploading a Confidential Methodology Section

The example illustrates why a limited upload is not necessarily a minor disclosure. A single methodology section may contain the central intellectual contribution of the proposal.

It also demonstrates that the reviewer's obligation is not simply to avoid publicly posting confidential information. The obligation includes following the authorized conditions under which that information may be processed.

Misconceptions About Confidential Grant Proposals and AI Uploads

How to Protect Confidential Grant Applications During Review

When reviewing someone else's unpublished application, begin with the assumption that its contents must remain within the authorized review process unless the funding organization specifies otherwise.

Before using any external technology, establish whether the intended processing is permitted. This includes AI summarization, methodological analysis, critique drafting, and other activities that require the proposal's contents.

The distinction between applicants and reviewers is especially important. An applicant may have authority to process portions of their own proposal under suitable conditions, while a reviewer generally cannot make the same decision about another researcher's confidential application.

The broader question of uploading an applicant's own confidential proposal to AI involves different authorization considerations.

Reviewers should also remember that using generative AI to evaluate research applications raises scientific judgment and fairness concerns even when the confidentiality requirements can be satisfied.

Confidentiality Checklist for Grant Reviewers Considering AI

Frequently Asked Questions About Uploading Confidential Grant Proposals to AI

Reviewer Access Does Not Authorize External AI Disclosure

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.