Is Undisclosed AI Use Considered Research Misconduct?

If You Use AI Without Disclosing It, Have You Committed Research Misconduct?

A researcher uses generative artificial intelligence to improve a manuscript, generate analytical code, summarize literature, or help interpret findings. The work is eventually submitted to a journal without mentioning the AI assistance.

Is that omission research misconduct? Does the answer change if AI only corrected grammar, rather than contributing to the analysis or writing substantial portions of the paper?

Researchers may encounter conflicting expectations. Some publishers require disclosure of particular AI uses, while institutions and funders may impose additional requirements. Meanwhile, formal research-misconduct definitions do not necessarily classify every reporting violation as misconduct.

The important distinction is between failing to comply with a disclosure requirement, misleading readers about how research was conducted, and satisfying the legal or institutional criteria for a formal misconduct finding.

Undisclosed AI Use Is Not Automatically Formal Research Misconduct

How AI Nondisclosure Relates to Research Integrity and Misconduct

What does research misconduct actually mean?

Research misconduct is a formal category defined by the legal, institutional, or funding framework governing particular research activities.

Under the United States Public Health Service framework administered by the Office of Research Integrity (ORI), research misconduct concerns fabrication, falsification, or plagiarism in proposing, performing, reviewing, or reporting research.

The framework distinguishes misconduct from honest error and differences of opinion. A formal finding also requires the applicable evidentiary and culpability standards to be met.

This means that a researcher does not automatically commit misconduct simply because an AI-related disclosure statement was omitted.

Nevertheless, institutions and other research organizations may adopt broader policies addressing misrepresentation, failure to follow research requirements, or other unacceptable practices.

The question of when AI use becomes a research-integrity problem is therefore broader than whether a particular act satisfies one formal misconduct definition.

Why do journals require disclosure of AI assistance?

AI disclosure requirements can help readers, editors, and reviewers understand how a manuscript or research output was produced.

Such information may be relevant when AI has contributed to drafting, data processing, image generation, literature analysis, or interpretation.

Disclosure also helps establish which activities were performed by researchers and which involved automated systems.

The International Committee of Medical Journal Editors (ICMJE) recommends that authors disclose the use of AI-assisted technologies in submitted work and explain how they were used.

The Committee on Publication Ethics (COPE) similarly emphasizes human accountability and transparency concerning AI tools in scholarly publishing.

These recommendations should not be interpreted as a universal requirement that every journal demands identical disclosures for every AI-assisted activity.

Researchers must verify the specific policy applicable to the intended publication.

Does every use of AI require disclosure?

Not necessarily. Disclosure requirements vary among journals, publishers, institutions, and research activities.

Some policies distinguish routine language correction from substantive AI assistance. Others require broader declarations covering AI-generated text, images, code, or analytical contributions.

For example, a journal may treat automated spelling correction differently from using a large language model to draft the discussion section or generate interpretations of research findings.

Researchers should avoid assuming that a disclosure practice accepted by one publisher will be accepted by another.

The appropriate question is not merely whether AI was used, but whether the particular activity falls within the governing disclosure requirements.

When does nondisclosure become misleading?

Nondisclosure becomes especially consequential when it prevents readers from understanding how important research activities were performed.

Suppose a manuscript states that two researchers independently coded qualitative interviews, but the actual coding was generated by an AI system and the researchers did not conduct the claimed independent coding.

The problem extends beyond the absence of an AI disclosure statement. The methods section inaccurately describes the analytical procedure.

Similarly, a researcher who claims to have manually examined every article in a systematic review while relying on unverified AI-generated classifications may misrepresent the research process.

Material misrepresentation can undermine the credibility of the findings even when the underlying data are genuine.

Whether the conduct meets a formal misconduct definition requires a separate assessment under the applicable framework.

How does nondisclosure differ from fabrication, falsification, and plagiarism?

These categories concern different forms of conduct and should not be collapsed into a single label.

AI involvement does not create a separate exemption from established misconduct standards.

Equally, the presence of an AI-related reporting error does not establish fabrication, falsification, or plagiarism without evidence supporting that conclusion.

Can undisclosed AI-generated text constitute plagiarism?

Not automatically.

Plagiarism generally concerns the appropriation of another person's ideas, processes, results, or words without appropriate credit under the applicable standard.

AI-generated text may contain material resembling existing publications, and researchers must verify sources and attribution where relevant.

However, failing to identify an AI tool is not necessarily equivalent to appropriating another human author's work.

The two concerns may overlap when AI output reproduces identifiable source material that is then presented without appropriate acknowledgment.

Researchers should distinguish disclosure of AI assistance from citation of the scholarly sources supporting their claims.

An AI disclosure statement does not replace citations, and proper citations do not necessarily satisfy a separate AI disclosure requirement.

Does undisclosed AI assistance affect authorship?

It can raise questions about whether the named authors accurately represent their contributions and accept responsibility for the work.

The ICMJE states that AI-assisted technologies should not be listed as authors because they cannot fulfill the responsibilities associated with authorship.

Human authors remain responsible for reviewing the accuracy, integrity, and originality of material generated with AI assistance.

However, using AI to support drafting does not automatically disqualify a human researcher from authorship.

Authorship eligibility depends on the relevant criteria and the individual's actual scholarly contributions.

The central concern is whether the human authors meet the applicable requirements and whether the use of AI has been reported as required.

What if the researcher did not know disclosure was required?

A lack of awareness may be relevant when determining responsibility or an appropriate institutional response, but it does not automatically eliminate a policy violation.

Researchers are generally expected to consult the submission requirements of the journals to which they submit their work.

However, formal misconduct determinations may require consideration of intention, knowledge, recklessness, and other circumstances.

An inadvertent omission corrected promptly after discovery should not automatically be treated as equivalent to deliberate concealment of prohibited AI use.

At the same time, researchers should not rely on claimed ignorance when a policy was clearly communicated and the omission materially misrepresented the work.

Can AI nondisclosure lead to manuscript rejection, correction, or retraction?

Potentially. Editorial responses depend on the publisher's policies, the seriousness of the omission, and whether the reliability of the research is affected.

A journal may request clarification or a corrected disclosure statement. More serious circumstances may warrant rejection, investigation, correction, or retraction.

Retraction is not necessarily a finding that the authors committed misconduct. The Committee on Publication Ethics distinguishes concerns about the reliability of the published record from formal determinations of individual culpability.

For example, an article containing fabricated AI-generated data may require retraction because its findings are unreliable, regardless of whether an institutional misconduct investigation has concluded.

By contrast, an omitted statement concerning otherwise permissible language assistance may be resolved through clarification or correction, depending on the journal's policy.

Should researchers disclose AI use even when a policy is silent?

Disclosure may still be appropriate when AI materially influenced the research process or interpretation.

A methods section should accurately describe consequential analytical procedures, regardless of whether the journal has a separate AI disclosure requirement.

For example, if a large language model performed an initial classification of research materials, the model's role may be methodologically relevant even when the journal does not explicitly require a standalone AI statement.

Researchers should consider what information another scholar would need to understand, evaluate, or reproduce the work.

However, transparency should be proportionate. An exhaustive record of every routine spelling suggestion is not necessarily useful or required.

The relevant standard is accurate reporting of material research activity alongside compliance with applicable policies.

Who determines whether nondisclosure constitutes misconduct?

Formal research-misconduct findings are generally made through the procedures established by the responsible institution or other authorized body.

Journal editors may determine whether a manuscript violates publication policies or whether the scholarly record requires correction.

These responsibilities overlap but are not identical.

An editor may reject a manuscript for violating AI disclosure requirements without establishing formal research misconduct.

Likewise, an institution may investigate alleged fabrication or falsification involving AI even when a journal has already issued an editorial correction.

Researchers should avoid making misconduct accusations based solely on an AI-detection score, stylistic suspicion, or the absence of a disclosure statement.

Reliable assessment requires evidence of the relevant conduct and application of the governing standards.

Two Researchers Fail to Disclose AI Assistance, but the Consequences Differ

Both researchers failed to disclose AI assistance, but the underlying conduct differs substantially.

The example illustrates why nondisclosure should be assessed in context rather than automatically equated with formal research misconduct.

Common Misconceptions About Undisclosed AI Use

Disclose Material AI Assistance and Correct Omissions Promptly

Researchers should approach AI disclosure as part of accurate scholarly reporting rather than as a statement that automatically determines whether the research is ethical.

The first responsibility is to ensure that the underlying research methods, evidence, and claims are truthful.

The second is to comply with the applicable disclosure requirements.

Researchers should also recognize that human accountability remains central when AI performs a prohibited research action.

The purpose of disclosure is not to transfer responsibility to AI. It is to provide an accurate account of how the work was produced.

Before Submitting Research That Used Generative AI

Frequently Asked Questions About Undisclosed AI Assistance

AI Nondisclosure May Be a Violation Without Automatically Being Misconduct

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.