NIH AI Policy

NIH AI policy is not a single rulebook but three separate NIH Guide notices that apply to three different actors at three different stages of the grant lifecycle. Confusing them is the most common mistake: a rule that binds a peer reviewer does not bind an applicant, and a rule that binds an applicant does not bind a funded investigator writing a progress report. This page lays out all three, what each one actually requires, and how NIH enforces them. Last verified: August 16, 2026.

Who it governs	Notice	Effective / released	The rule in one line
Peer reviewers	NOT-OD-23-149	June 23, 2023	Prohibited from using generative AI to analyze applications or draft critiques
Applicants	NOT-OD-25-132	Applications from Sept. 25, 2025 receipt date on	AI-developed content isn’t treated as an original idea; 6-application/PI/year cap
Funded researchers	May 2026 Extramural Nexus notice	May 2026	Existing FFP misconduct rules apply to AI-related lapses in conduct/reporting

1. Peer reviewers: generative AI is prohibited outright (NOT-OD-23-149)

NIH’s oldest and strictest AI rule applies to its own scientific peer reviewers. NOT-OD-23-149, released June 23, 2023, prohibits NIH peer reviewers from using natural language processors, large language models, or other generative AI technologies to analyze grant applications or R&D contract proposals, or to help formulate their written critiques. The notice frames this as a confidentiality and integrity issue, not a quality-of-review one: uploading any part of an application, proposal, or critique to an online generative AI tool discloses confidential, unpublished content to a third party with no guarantee of where that data is stored, viewed, or reused — a direct violation of NIH’s peer review confidentiality and nondisclosure requirements, which the notice was issued specifically to clarify (it amends the earlier NOT-OD-22-044 on peer review security and confidentiality).

Mechanically, NIH enforces this through paperwork every reviewer already signs: as part of standard pre-meeting certification, reviewers must sign a modified Security, Confidentiality, and Nondisclosure Agreement that explicitly names this prohibition. The same policy extends to members of NIH’s National Advisory Councils and Boards. There is one narrow carve-out: assistive technologies used for accessibility needs may be granted an exception, but only if the reviewer discloses the specific technology to the Scientific Review Officer or other designated NIH official before the review meeting.

2. Applicants: AI-written proposals aren’t treated as original, and there’s a new application cap (NOT-OD-25-132)

NOT-OD-25-132, “Supporting Fairness and Originality in NIH Research Applications,” released July 17, 2025, addresses the other side of the process: what happens when applicants — not reviewers — use AI. NIH states plainly that it will not consider applications that are either substantially developed by AI, or that contain sections substantially developed by AI, to be original ideas of the applicant. This is a direct application of the existing originality expectation in NIH Grants Policy Statement Section 2.1.2, which requires institutions and research teams to propose original ideas for funding — NOT-OD-25-132 doesn’t create a new originality standard, it clarifies that AI-generated content doesn’t satisfy the existing one.

The notice pairs that originality rule with a new, separate limit: NIH will accept no more than six new, renewal, resubmission, or revision applications from a single Principal Investigator, Program Director, or Multiple-PI, across all council rounds, in a calendar year. T activity codes and R13 conference grant applications are excluded from the cap. NIH’s stated rationale, in the notice’s own background section, is that AI tools had enabled some PIs to submit more than 40 distinct applications in a single submission round, straining NIH’s review capacity even though the overall share of PIs submitting more than six applications a year is small. NIH also states it will continue to use AI-detection technology on submitted applications. The cap applies to applications submitted to the September 25, 2025 receipt date and beyond.

Consequences described in the notice are severe and not contingent on a completed misconduct finding: if AI-generated content is identified after an award has been made, NIH may refer the matter to the Office of Research Integrity (ORI) to determine whether it rises to research misconduct — while separately and simultaneously pursuing grants-management enforcement, including disallowing costs, withholding future awards, suspending the grant in whole or in part, or terminating it outright.

3. Funded researchers: AI-related fabrication and plagiarism are evaluated under the existing misconduct standard (May 2026 notice)

In May 2026, NIH’s Office of Extramural Research used its Extramural Nexus newsletter to extend the same underlying logic past the application stage, into the conduct and reporting of research NIH has already funded. The notice, “Helpful Reminders to Ensure Integrity of NIH-Supported Research When Using Artificial Intelligence,” doesn’t create a new misconduct category for AI — it ties AI use directly to the existing federal research misconduct definition (fabrication, falsification, and plagiarism, codified at 42 CFR Part 93) and names two AI-specific scenarios explicitly: presenting AI-hallucinated, non-existent citations as real (fabrication), and using an AI tool to reproduce or closely paraphrase someone else’s published work without disclosure or attribution (plagiarism).

The operative standard is the same one that already governs any other FFP finding: a researcher crosses into misconduct when they use AI tools intentionally, knowingly, or recklessly in ways that deviate from accepted research practice. NIH is explicit that the burden of verification sits with the researcher — “the AI got it wrong” is not, on its own, a defense; failing to check AI output before submitting it is the lapse being flagged. Read together with the other two notices, NIH’s position spans the entire award lifecycle: reviewer conduct (NOT-OD-23-149), application content (NOT-OD-25-132), and the conduct and reporting of the funded work itself (the May 2026 reminder). CASRAI has a full breakdown of the May 2026 notice, including its enforcement mechanics, in NIH’s May 2026 Notice: When AI Use Becomes Research Misconduct.

What NIH does not currently require

Unlike the NSF AI policy, NIH has not issued a general notice encouraging or requiring applicants to disclose generative-AI use when preparing a proposal — NIH’s application-side rule is an originality standard (AI-substantially-developed content isn’t treated as original) plus the six-application annual cap, not a disclosure checkbox. NIH also has not published a blanket prohibition on using AI tools anywhere in the conduct of funded research; the May 2026 notice is explicit that AI-assisted drafting, summarizing, and literature search are not themselves the problem it targets. Treat any of these as subject to change — NIH has issued three separate AI-related notices in under three years, and its own May 2026 notice was framed as a “reminder” building on the two before it, which suggests more may follow. Always check the current NIH Guide for Grants and Contracts notices directly before relying on this summary for a live submission or review assignment.

How NIH compares to other funders

NIH’s three-notice, lifecycle-staged approach is one of several distinct funder positions on generative AI. NSF instead encourages (but does not yet require) proposal-preparation AI disclosure and folds AI-assisted misconduct into its existing PAPPG misconduct definition; the European Research Council frames the issue differently again. See NIH vs NSF vs ERC: AI Policies for Grant Writing for a side-by-side comparison, and AI in peer review for how the reviewer-side prohibition compares to publisher peer-review AI policies.

Related terms

• NSF AI Policy
• AI in peer review
• Research misconduct
• ORI (US Office of Research Integrity)
• Generative-AI disclosure statement
• NIH Grants Policy Statement (NIH GPS)

Frequently Asked Questions

Can NIH peer reviewers use ChatGPT or other AI tools to write their critiques?

No. NOT-OD-23-149, released June 23, 2023, prohibits NIH scientific peer reviewers from using natural language processors, large language models, or other generative AI technologies to analyze grant applications or R&D contract proposals, or to help formulate critiques. Uploading any part of an application or critique to an online AI tool violates NIH’s peer review confidentiality and nondisclosure requirements. Reviewers certify compliance with this prohibition as part of standard pre-meeting paperwork, and only a narrow, pre-disclosed accessibility-technology exception applies.

Does NIH allow applicants to use AI to help write a grant application?

NIH does not prohibit using AI as a drafting aid, but under NOT-OD-25-132 (July 17, 2025), it will not treat an application that is substantially developed by AI, or that contains sections substantially developed by AI, as an original idea of the applicant — and originality is a standing requirement under NIH Grants Policy Statement Section 2.1.2. NIH also states it uses AI-detection technology on submitted applications and may pursue both a research-misconduct referral to ORI and independent grants-management enforcement if AI-generated content is identified after an award is made.

What is the NIH six-application cap and does it apply to everyone?

NOT-OD-25-132 limits any single Principal Investigator, Program Director, or Multiple-PI to six new, renewal, resubmission, or revision applications across all council rounds per calendar year, effective for applications submitted to the September 25, 2025 receipt date and beyond. T activity codes and R13 Conference Grant applications are excluded. NIH introduced the cap after observing PIs submit more than 40 applications in a single round, which it attributed in part to AI-assisted application generation.

Is using AI to draft a progress report or manuscript automatically research misconduct at NIH?

No. NIH’s May 2026 Extramural Nexus notice states that a researcher crosses into research misconduct only when they use AI intentionally, knowingly, or recklessly in ways that deviate from accepted research practice — for example, submitting a fabricated AI-generated citation without checking it, or presenting AI-paraphrased text as original without disclosure or attribution. AI-assisted drafting, summarizing, or literature search, with proper verification and disclosure, is not what the notice targets.

Does NIH require applicants to disclose when they used generative AI, the way NSF does?

Not currently. NIH’s application-side rule under NOT-OD-25-132 is an originality standard and an application-volume cap, not a disclosure requirement. This differs from NSF, whose December 2023 notice encourages (though does not yet mandate) proposers to disclose generative-AI use in the Project Description. See NSF AI Policy for the comparison.