In May 2026, NIH’s Office of Extramural Research used its Extramural Nexus newsletter to spell out, in more direct terms than it had before, exactly where the use of generative AI tools in NIH-supported research can cross the line into research misconduct. The notice, titled “Helpful Reminders to Ensure Integrity of NIH-Supported Research When Using Artificial Intelligence”, does not create new policy so much as connect existing rules — the federal research misconduct definition, NIH’s application-integrity notice NOT-OD-25-132, and the peer-review AI prohibition in NOT-OD-23-149 — to the everyday reality of investigators and staff now routinely using AI tools to draft, summarize, code, and search literature.
What the notice actually says
NIH opens by acknowledging the obvious: AI adoption has produced real advances in biomedical research, from literature triage to data analysis. But it pairs that acknowledgment with a warning that AI use has also introduced new ways for the underlying integrity of NIH-supported research to be compromised — not through some novel category of wrongdoing, but through AI making old failure modes easier to commit, sometimes without the researcher fully realizing it.
The notice ties this directly back to the federal research misconduct definition — fabrication, falsification, and plagiarism (FFP) — codified for HHS-funded research at 42 CFR Part 93. Two AI-specific scenarios get called out explicitly:
- Fabricated or hallucinated citations presented as real. Large language models routinely generate plausible-looking references to papers, datasets, or findings that do not exist. NIH’s position is that presenting AI-generated, non-existent references as genuine can itself constitute data fabrication under the misconduct definition — the fact that a machine produced the false content rather than the researcher typing it manually does not change the classification.
- Undisclosed reuse of text via AI tools. Using an AI tool to paraphrase or reproduce substantial portions of someone else’s published work without disclosure or attribution can constitute plagiarism in the same way manual copying would. The notice frames the AI tool as a means of committing a familiar violation, not a defense against one.
Where NIH draws the misconduct line
The notice’s most operationally useful sentence is its statement of intent: researchers cross into research misconduct territory when they intentionally, knowingly, or recklessly use AI tools in ways that deviate from accepted research practices. That three-part standard — intentional, knowing, or reckless — is the same culpability threshold that already governs FFP findings generally; NIH is not lowering the bar for AI-related conduct, it is clarifying that AI-assisted fabrication or plagiarism is evaluated under the identical standard as any other kind.
Practically, this means an investigator who unknowingly submits an AI-hallucinated citation because they failed to check it is in a materially different — though still serious — position than one who knowingly lets a fabricated reference stand after being alerted to it. The notice puts the verification burden squarely on the researcher: NIH states that researchers remain accountable for the validity of data and citations that come out of an AI tool, regardless of which tool produced them. “The AI got it wrong” is not, on its own, a defense; failing to verify AI output before relying on it or submitting it is itself the lapse NIH is warning about.
What NIH says researchers and institutions should do
The notice’s practical guidance is consistent with what AI disclosure norms elsewhere in scholarly publishing already require: transparent disclosure of where and how AI tools were used, and proper attribution of any material the tools helped produce. For grant applications specifically, this sits alongside — and does not replace — NIH’s separate, earlier notice on application content, NOT-OD-25-132, “Supporting Fairness and Originality in NIH Research Applications” (July 2025), which states that NIH will not treat applications substantially developed by AI, or containing sections substantially developed by AI, as original ideas of the applicant. The May 2026 reminder extends that same expectation of originality and verification past the application stage into the conduct and reporting of awarded research itself.
It also sits alongside NOT-OD-23-149, which prohibits NIH scientific peer reviewers from using generative AI tools to analyze applications or draft critiques, on confidentiality grounds. Read together, the three notices cover the full award lifecycle NIH is concerned about: peer review (NOT-OD-23-149), application preparation (NOT-OD-25-132), and the conduct and reporting of the funded research itself (the May 2026 reminder).
Enforcement: what happens if AI misuse is identified
The notice describes two tracks that can run simultaneously rather than sequentially. If AI-related misconduct is identified after an award has been made, NIH may refer the matter to the Office of Research Integrity (ORI) to determine whether a formal finding of research misconduct is warranted — the same FFP inquiry/investigation process described in NIH’s broader research integrity and misconduct guidance. At the same time, NIH states it may pursue grants-management enforcement independent of that ORI determination, including disallowing costs, withholding future awards, suspending the grant in whole or in part, and terminating the award. In other words, an institution does not need to wait for ORI to conclude a misconduct inquiry before NIH can act on the funding side — the two processes are described as parallel, not gated on each other.
What it means for NIH-funded researchers and institutions
For working researchers, the practical takeaway is unglamorous but concrete: AI-assisted drafting, summarizing, or literature searching is not itself the problem NIH is flagging — unverified AI output making it into an application, progress report, or publication is. Every citation an AI tool surfaces needs to be checked against a real source before it is submitted; every substantial block of AI-assisted text needs the same attribution scrutiny a researcher would apply to any other secondary source.
For research administration and compliance offices, the notice is a signal to fold AI-specific language into existing research integrity training rather than standing up a separate track — it reinforces, rather than replaces, the RCR and misconduct-process training many institutions already require of NIH senior/key personnel. Institutions that have not yet updated their responsible conduct of research materials to address AI-assisted fabrication and plagiarism explicitly now have a directly citable NIH source for doing so.
The notice also arrives against a wider funder-level backdrop: NSF issued its own AI Dear Colleague Letter addressing responsible AI use in the proposal and merit-review process, and journals and publishers have been converging on disclosure-based AI policies for several years, covered in CASRAI’s AI disclosure guide. NIH’s May 2026 reminder is best read as the research-conduct-and-reporting piece of that same broader shift — funders and publishers are no longer treating “we didn’t have a rule for AI yet” as a live option; they are applying existing integrity, originality, and attribution rules to AI-assisted work and saying so explicitly.
Frequently asked questions
Does the NIH notice ban the use of AI tools in NIH-funded research?
No. The notice does not prohibit AI tool use in the conduct of NIH-supported research generally — it addresses application content (via NOT-OD-25-132) and peer review (via NOT-OD-23-149) with specific restrictions, but for the broader conduct and reporting of awarded research, its focus is on verification, disclosure, and attribution rather than a blanket ban.
Is using AI to draft a grant application or manuscript automatically research misconduct?
No. NIH’s standard is that misconduct occurs when a researcher intentionally, knowingly, or recklessly uses AI in ways that deviate from accepted research practices — for example, submitting a fabricated AI-generated citation without checking it, or presenting AI-paraphrased text as original without disclosure. Using AI as a drafting aid, with verification and appropriate disclosure, is not what the notice targets.
Who investigates AI-related research misconduct allegations at NIH-funded institutions?
The same body that investigates any other FFP allegation involving PHS-funded research: the institution’s own research integrity officer conducts the inquiry and investigation under 42 CFR Part 93, with findings reportable to the HHS Office of Research Integrity. NIH’s grants management staff can act on funding separately and in parallel.
How does this notice relate to NOT-OD-25-132?
NOT-OD-25-132 (July 2025) addresses AI use at the application stage, stating NIH will not treat content substantially developed by AI as the applicant’s original idea. The May 2026 notice is a broader reminder that extends similar expectations of verification, disclosure, and attribution to the conduct and reporting of already-funded research, and explains how AI-related lapses map onto the existing FFP misconduct framework.
