The Office of Research Integrity (ORI) released its “Guidance on Generative Artificial Intelligence” on August 14, 2026.[1] The guidance does not create a new category of artificial intelligence (AI) misconduct, nor does it alter the legal standard for research misconduct under 42 CFR Part 93.[2] Instead, it applies the existing framework to a rapidly evolving technological environment and, in doing so, highlights a new readiness challenge for research institutions. Institutions may increasingly need the policies, expertise, documentation practices, and evidentiary capabilities necessary to evaluate AI-related conduct under standards that were not written specifically for AI.
The significance of the guidance lies less in what it changes legally than in what it changes operationally. A finding of research misconduct still requires a significant departure from accepted practices of the relevant research community, intentional, knowing, or reckless conduct, and proof by a preponderance of the evidence. The burden remains on institutions or the U.S. Department of Health and Human Services to establish those elements.
The challenge is that AI introduces new questions regarding accepted practice, evidence preservation, scientific oversight, and investigative methodology. The ORI’s technology-neutral approach leaves institutions responsible for applying existing standards to evolving technologies, disciplinary norms, publication requirements, and sponsor expectations. As a result, institutions that have not yet considered how AI use will be documented, reviewed, and assessed may find themselves unprepared when allegations arise.
Although ORI does not require institutions to adopt AI-specific misconduct policies, the guidance suggests that institutional policies addressing AI use may assist committees evaluating research misconduct allegations. The practical implication is that organizations increasingly need a coordinated framework for acceptable AI use, disclosure, verification, data protection, and evidence preservation, even if the misconduct policy itself remains unchanged.
What ORI Clarifies
AI detection is a lead, not a verdict. ORI makes clear that AI-enabled detection tools should not serve as the sole basis for a research misconduct finding. Automated tools may generate investigative leads or identify anomalous patterns, but committees must evaluate the full evidentiary record and determine whether the research was conducted as reported. ORI further notes that investigation committees must include appropriate scientific expertise and may benefit from additional AI subject-matter expertise when AI detection tools are used or when AI-related issues are central to the matter under review.
The practical consequence is significant. Institutions should view AI-generated indicators the same way they view other forms of preliminary evidence: as a starting point for inquiry rather than a substitute for analysis. Ultimately, the strength of a case will depend not on the AI-detection signal itself, but on the institution’s ability to corroborate it through research records, source materials, witness testimony, forensic analysis, and other objective evidence that can support a defensible determination.
Plagiarism Detection May Be Less Effective in AI-Assisted Content
ORI cautions that plagiarism detection tools may have limited ability to identify plagiarized material contained within AI-generated scientific content such as grant proposals, reports, or manuscripts. Consequently, allegations involving AI-assisted writing may require more extensive review than traditional plagiarism investigations.
The guidance also reminds researchers that AI-generated text or ideas may reproduce content originating from another person’s work. Researchers therefore remain responsible for evaluating whether generated content incorporates protected intellectual contributions and whether attribution is required.
Disclosure Helps, But It Is Not a Safe Harbor
ORI encourages researchers to identify and explain the use of generative AI tools during research, including grant proposal and manuscript preparation. According to the guidance, transparent disclosure may support reproducibility and may provide evidence relevant to a defense against certain misconduct allegations. At the same time, disclosure does not preclude a misconduct finding. AI use that departs significantly from accepted practices may still support a finding of misconduct, while undisclosed or inaccurately described AI-based data generation or processing may raise fabrication or falsification concerns.
This distinction is important because many organizations are focused on disclosure requirements alone. Transparency is increasingly important, but disclosure should be viewed as one component of a broader governance framework rather than a complete solution.
Hallucinations Create Classification Challenges
One of the more nuanced portions of the guidance involves fabricated citations. ORI explains that fabrication under 42 CFR Part 93 concerns data or results. When references do not constitute data or results, a hallucinated citation may fall outside the regulatory definition of fabrication. However, in contexts such as literature reviews where references themselves form part of the research data, hallucinated citations could be classified as fabrication.
Similarly, AI-generated text or ideas may support plagiarism allegations if they appropriate another person’s work without appropriate credit. The key issue is not whether AI was involved, but whether the underlying conduct satisfies the established misconduct framework.
Authorship Remains a Separate Issue
The guidance does not establish a new AI-authorship rule. Moreover, 42 CFR Part 93 expressly excludes authorship and credit disputes from the definition of plagiarism. Institutions therefore should not assume that questions about AI-assisted authorship should automatically be routed through the research misconduct process. Instead, AI-related authorship concerns should generally be addressed through institutional publication, authorship, and accountability policies, while separately evaluating whether any related conduct implicates fabrication, falsification, or plagiarism.
Sponsor Policies May Matter as Much as ORI Guidance
An important feature of the revised research misconduct framework is that accepted practices are informed not only by general scientific norms, but also by Public Health Service funding-component policies. As agencies continue to develop AI-specific expectations, those requirements may increasingly influence how institutions evaluate accepted practice in specific research communities.
The National Institutes for Health (NIH) has already established significant expectations regarding AI use. NIH has stated that applications substantially developed by AI may not be considered the applicant’s original ideas and may be referred to ORI if identified after award, while also triggering separate enforcement measures.[3] NIH also prohibits scientific peer reviewers from using generative AI to analyze applications or prepare critiques because doing so may compromise peer-review confidentiality obligations.[4]
As a result, conduct that falls short of the federal definition of research misconduct may nevertheless violate other federal or sponsor requirements, institutional standards, or publication policies. Conversely, a violation of an institutional AI policy should not be presumed to constitute research misconduct.
Where the Guidance Leaves Institutions With Difficult Decisions
ORI intentionally adopted a flexible, technology-neutral framework. That flexibility is valuable because AI technologies and scientific practices are changing rapidly. However, flexibility also means institutions retain responsibility for making difficult judgment calls. The guidance does not prescribe a universal disclosure framework, specify which AI-generated materials should be retained, establish a reliability threshold for AI-detection tools, or resolve how institutions should evaluate evolving disciplinary norms.
Perhaps the most immediate operational challenge involves evidence preservation. Research misconduct regulations require institutions to secure relevant research records and retain the institutional record. Yet, AI-related evidence may reside across personal devices, third-party AI platforms, browser sessions, transient chat histories, collaborative applications, and cloud-based environments that may be difficult to identify and recover after an allegation emerges.
The question is no longer simply whether an institution can assess AI-related conduct. It is whether the institution can locate, preserve, authenticate, and explain the evidence necessary to support a defensible conclusion.
5 Takeaways for Research Institutions and Research Integrity Officers
1. Do Not Treat an AI-detection Score as a Finding
Use automated tools to identify potential issues and investigative leads, but rely on comprehensive evidence gathering and analysis before reaching conclusions.
2. Develop a Risk-Based AI Evidence Preservation Protocol
Institutions should determine when potentially relevant prompts, outputs, uploaded materials, system metadata, model information, and intermediate processing files should be preserved following an allegation or anticipated investigation.
3. Maintain a Sponsor-Policy Inventory
Track NIH and other funding-agency requirements relating to AI use, disclosure, confidentiality, peer review, and data controls because those requirements may influence accepted-practice determinations and create consequences independent of Part 93.
4. Separate Misconduct Allegations From Broader AI Compliance Concerns
Not every AI issue is a research misconduct matter. Institutions should distinguish potential fabrication, falsification, and plagiarism allegations from concerns involving publication standards, confidentiality obligations, data governance, privacy, employment policies, or sponsor requirements.
5. Establish a Defensible Institutional Baseline
Clear guidance regarding acceptable use, disclosure expectations, human verification, attribution, data protection, and documentation can help researchers act consistently and provide misconduct committees with a more reliable reference point when evaluating departures from accepted practice.
A Readiness Test for Research Institutions
ORI’s guidance should be viewed as more than an AI policy update. It is a reminder that research integrity oversight increasingly depends on an institution’s ability to apply longstanding principles in unfamiliar technological environments. The organizations best positioned to manage AI-related integrity risks will not necessarily be those with the most restrictive AI policies. They will be the institutions that can clearly articulate acceptable practices, preserve relevant evidence, distinguish misconduct from other forms of noncompliance, and support defensible decisions when questions arise.
In the next phase of research integrity oversight, institutional credibility may depend less on whether AI is permitted and more on whether organizations can explain, document, and defend how AI was used.
Sources
[1] The Office of Research Integrity, Guidance for Public Health Service Policies on Research Misconduct Generative Artificial Intelligence, August 2026.
[2] 42 CFR Part 93 Public Health Service Policies on Research Misconduct
https://www.ecfr.gov/current/title-42/chapter-I/subchapter-H/part-93
[3] National Institutes of Health Notice NOT-OD-25-132, “Supporting Fairness and Originality in NIH Research Applications,” July 17, 2025. https://grants.nih.gov/grants/guide/notice-files/NOT-OD-25-132.html
[4] National Institutes of Health Notice NOT-OD-23-149, “The Use of Generative Artificial Intelligence Technologies is Prohibited for the NIH Peer Review Process,” June 23, 2023. https://grants.nih.gov/grants/guide/notice-files/NOT-OD-23-149.html
© Copyright 2026. The views expressed herein are those of the author(s) and not necessarily the views of Ankura Consulting Group, LLC, its management, its subsidiaries, its affiliates, or its other professionals. Ankura is not a law firm and cannot provide legal advice.
