NIH Bars Generative AI in Peer Review and Limits AI-Developed Grant Applications

On 23 June 2023 NIH issued notice NOT-OD-23-149, which prohibits NIH scientific peer reviewers from using natural language processors, large language models or other generative AI technologies for analysing and formulating peer review critiques for grant applications and research and development contract proposals. The prohibition also covers members of NIH National Advisory Councils and Boards.

The stated reason is confidentiality. NIH notes that AI tools offer no guarantee of where data are being sent, saved, viewed or used in the future, and that uploading application content to an online generative AI tool violates NIH’s peer review confidentiality and integrity requirements.

As of November 2026, NIH’s peer review policies page still lists NOT-OD-23-149 as a current policy notice. Pubrica found no NIH notice superseding it. This contrasts with some other funders that relaxed their reviewer rules in 2026, such as the Australian Research Council and NHMRC, which now allow limited reviewer use under conditions.

On 17 July 2025 NIH issued NOT-OD-25-132, Supporting Fairness and Originality in NIH Research Applications. It states that applications, or sections of applications, substantially developed by AI will not be considered the original ideas of the applicants. According to university summaries of the notice, NIH may not consider applications that are substantially developed by AI or contain AI-generated content of concern, and may take action where such use is identified after funding, including referral for research misconduct review.

The same notice introduced a limit of six applications per principal investigator, applying to applications submitted for the 25 September 2025 deadline and beyond. New, renewal, resubmission and revision applications count toward the limit. NIH explained in its Extramural Nexus that the limit responded to a small number of investigators submitting far more applications than NIH traditionally expects. Sources differ on whether the cap and the AI provisions should be treated as separate measures; they appear in the same notice.

The notice does not prohibit all AI use in preparing applications. Some university guidance describes AI assistance as not banned, but because NIH has not published a threshold for what counts as substantially developed by AI, research offices generally advise caution.

In May 2026 NIH published an Extramural News item, described by universities as Helpful Reminders to Ensure Integrity of NIH-Supported Research When Using Artificial Intelligence. University research offices date it to 14 May 2026. According to secondary summaries, it does not create new rules but reminds the community that the peer review prohibition, the application originality policy and the federal research misconduct definition together apply across the award lifecycle, from review to the conduct and reporting of funded research.

Summaries report that NIH can pursue a research misconduct review, refer a case to the HHS Office of Research Integrity, and take enforcement action such as disallowance of costs or termination of an award where AI misuse is identified, including after funding has begun. Pubrica could not access the NIH text directly, so institutions should read the original item for exact wording.

Reviewers should keep all application materials out of generative AI tools, including for grammar checking or summarising, and should prepare critiques themselves. Applicants should ensure that the scientific ideas, aims and approach are their own, keep drafts and records showing how the application was developed, and use AI only for limited support consistent with institutional guidance.

NIH’s position is now stricter than that of several other major funders. In the United Kingdom, a revised joint statement by UKRI, Wellcome and other funders in June 2026 narrowed the reviewer rule to a prohibition on inputting confidential content into generative AI tools, and in Australia the ARC and NHMRC have, since 28 April 2026, allowed reviewers to use AI to refine the wording of their comments. Researchers who review for more than one funder should not assume that a practice permitted elsewhere is acceptable for NIH review.

Research offices and grant writing teams should track the number of applications each investigator submits to stay within the six-application limit, and should update internal guidance on AI use in proposal development. Medical writers supporting NIH applications should agree with investigators in advance how any AI tools will be used and documented. This brief summarises public NIH notices for information and does not represent NIH.