01 · The Question
Can Your Funder Tell You How You May Use AI?
Your university permits generative AI for research. You use it to brainstorm a grant proposal, refine some text, analyse data, or prepare a project report. Does institutional permission settle the matter?
Not necessarily. Funding organisations can establish conditions governing applications, assessment processes, funded activities, reporting, research integrity, confidentiality, data management, and other aspects of awards. Increasingly, some funders are also publishing explicit rules for generative AI.
These policies are not uniform. One funder may permit certain AI-assisted application activities with transparency requirements, while another may reject applications containing sections substantially developed by AI. What matters is the policy of the funder and funding opportunity that actually applies to you.
03 · What You Need to Know
Funder AI Rules Can Apply at Several Stages of Research
Funding Applications Are Already Being Regulated
One of the clearest areas of emerging funder policy concerns the preparation of grant applications. Funders have legitimate interests in originality, research integrity, confidentiality, fairness, and whether the proposal genuinely represents the applicants' ideas and capabilities.
Current policies show how differently funders can approach the same technology.
UK Research and Innovation permits generative AI to be used during application preparation subject to defined responsibilities. Its current policy expects transparency for substantive uses such as assisting idea generation, developing hypotheses, analysing or summarising information, comparing literature, generating or refining code, or generating an abstract. Minimal uses such as improving English, translation, formatting, or reducing word count do not need to be disclosed under that policy. UKRI also states that applicants must not use generative AI to generate an entire application or sections of an application without human involvement.
The U.S. National Institutes of Health takes a different approach to substantial AI generation. For applications submitted for the September 25, 2025 receipt date and beyond, NIH states that applications or sections substantially developed by AI will not be considered the applicants' original ideas. NIH warns that identification of such use after an award can lead to enforcement action.
The U.S. National Science Foundation provides another model. NSF encourages proposers to indicate the extent to which generative AI was used and how it contributed to proposal development, while separately imposing strict confidentiality requirements on its reviewers.
| Funder Example |
Current Approach |
Practical Lesson |
| UKRI |
Allows AI-assisted application preparation within specified responsibilities; expects transparency for substantive use and distinguishes it from minimal use. |
Permitted AI assistance can still carry integrity, confidentiality, and transparency conditions. |
| NIH |
Does not consider applications or sections substantially developed by AI to represent applicants' original ideas. |
Extensive AI generation can affect application eligibility even if ordinary writing assistance is permitted elsewhere. |
| NSF |
Encourages proposers to indicate AI use in proposal development and imposes strong confidentiality restrictions during merit review. |
Rules for applicants and reviewers may differ even within the same funding system. |
These examples should not be converted into a general rule such as "funders allow AI if you disclose it." They demonstrate precisely the opposite: different funders can draw the line differently.
Application Rules and Funded-Research Rules Are Not the Same Thing
A policy governing proposal preparation does not necessarily answer what happens after an award is made.
The funding application and the funded project are different stages. A funder might regulate AI-generated application text because originality matters to competitive assessment while permitting AI as an explicitly proposed research method. Conversely, a funder may allow limited AI assistance in writing an application while imposing stricter conditions on participant data, confidential information, security, research integrity, or project reporting after funding begins.
Read the scope of the policy. If it says that it governs "application preparation and assessment," do not silently extend it to every AI use during the resulting project. Equally, do not assume that permission during application preparation grants permission throughout the award.
Funders Can Restrict What Information Goes Into AI Systems
Confidentiality is particularly important in funding processes. Grant applications may contain unpublished hypotheses, preliminary findings, commercially valuable ideas, intellectual property, personal information, and detailed future research plans.
UKRI warns applicants to exercise caution when entering information into generative AI systems and states that sensitive or personal data belonging to others must not be entered without formal consent. Its policy also requires assessors not to enter any part of an application into generative AI while refining review language.
NSF similarly states that reviewers are prohibited from uploading proposal content, review information, and related records into non-approved generative AI tools. NSF treats such disclosure as incompatible with the confidentiality obligations of its merit-review process.
Watch Out
If a funder gives you confidential material as a reviewer or panel member, do not assume that your university's approved AI account makes uploading that material permissible. Reviewer confidentiality is an obligation created by the funding process itself. Check the funder's rules before using any AI system with application or review material.
Funders Can Treat Applicants and Reviewers Differently
This distinction is easy to miss. The same funder may permit applicants to use generative AI in limited ways while sharply restricting reviewers from doing so.
There is a sensible reason for the difference. Applicants control their own proposal material, subject to funder rules and the rights of collaborators. Reviewers receive someone else's confidential intellectual work under an obligation to protect it.
Under UKRI's current policy, for example, assessors may use generative AI only for language refinement and must not provide the system with any part of the application, personal information from it, or instructions to understand, summarise, or evaluate its content.
When reading a funder policy, therefore, identify your role. Applicant, principal investigator, research administrator, reviewer, panel member, and funded researcher may be subject to different provisions.
Funders Can Impose Requirements Through Award Conditions
Not every restriction needs to appear in a document titled "AI Policy." Funders can govern research through funding-opportunity conditions, grant agreements, award terms, research-integrity requirements, data-management obligations, security requirements, ethics conditions, and reporting rules.
This parallels the broader problem of multiple policies governing AI use in academic research. The relevant obligation may be embedded in an existing research rule rather than labelled as an AI requirement.
For example, an award condition restricting the transfer of sensitive project data to unauthorised third parties may constrain use of an external AI service even if the condition never mentions generative AI.
AI Can Be the Subject of the Funded Research Without Being Unrestricted
Funding a project that develops or evaluates AI obviously requires AI use. That does not mean every AI practice within the project is automatically acceptable.
The project may still be subject to conditions concerning research participants, data governance, security, intellectual property, reproducibility, ethics, responsible research and innovation, or the handling of sensitive technologies. What the funder agreed to support should be distinguished from how researchers are permitted to execute that work.
Your University and Funder Can Both Have Authority
Funder requirements do not normally replace institutional requirements. A funded researcher may need to comply with both.
Suppose your university allows generative AI for proposal drafting, but your intended funder prohibits sections substantially developed by AI. Institutional permission does not remove the funder's application condition.
The reverse matters too. A funder may allow an AI-assisted activity while your university prohibits uploading confidential research information to that service. Funder permission does not eliminate the institutional restriction.
Before reaching the funder layer, researchers should therefore know how to identify their own university's AI requirements.
Funding Policies Can Change During a Project
Funder AI policies are developing rapidly. UKRI's generative AI application and assessment policy, for example, was updated in July 2026, and the organisation announced further changes to its grant-assessment approach in September 2026 as it responded to increased use of generative AI.
For researchers, the practical lesson is not to memorise today's wording. Identify the authoritative policy page and determine which version or conditions apply to your application or award.
Long-running projects may need a process for monitoring changes in applicable AI policies rather than relying indefinitely on the requirements that existed when the proposal was submitted.
07 · A Quick Checklist
What to Check Before Using AI in Funded Research
Before using AI in a funding-related activity, check:
Find the funder's current official AI guidance rather than relying on another researcher's recollection.
Check the specific funding opportunity because it may contain additional requirements.
Determine whether you are subject to rules for applicants, reviewers, funded researchers, or more than one role.
Identify whether the funder distinguishes minimal AI assistance from substantive content generation.
Check whether substantive AI use must be disclosed and what information the funder expects.
Do not enter confidential application, reviewer, participant, or collaborator information into an AI service unless that processing is permitted.
For an awarded project, review the award conditions, approved proposal, data-management obligations, and relevant research-integrity requirements.
Verify that university, ethics, contractual, and other applicable requirements are also satisfied.
Recheck the authoritative policy when preparing a new application or entering a new stage of a long-running project.
