Is Undisclosed AI Use an Authorship Issue, Research-Integrity Issue, or Policy Violation?

What Kind of Research-Integrity Problem Is Undisclosed AI Use?

A researcher uses generative AI to revise a manuscript, develop part of its discussion, or assist with data analysis. The resulting paper is submitted without mentioning the AI assistance. Later, the journal discovers how the tool was used.

Has the researcher misrepresented authorship? Violated a disclosure policy? Committed research misconduct? Or could the omission be relatively minor if the AI only corrected grammar?

These classifications matter because they concern different obligations and may lead to different editorial or institutional responses. An undisclosed use of AI is not necessarily harmless, but neither does every omission justify the same allegation.

Undisclosed AI Use Can Involve More Than One Integrity Concern

How Authorship, Transparency, Policy Compliance, and Research Misconduct Intersect

Why is nondisclosure not one single category of misconduct?

Academic publishing involves several overlapping obligations. Authors must represent their contributions accurately, comply with submission requirements, acknowledge intellectual sources, and ensure that reported findings are trustworthy.

Undisclosed AI assistance can affect one or several of these obligations. The appropriate classification depends on the underlying conduct rather than the mere presence of an AI tool.

These categories are not mutually exclusive. A false AI declaration might accompany fabricated results or unattributed borrowing. Yet nondisclosure alone does not establish that those additional acts occurred.

When is undisclosed AI use primarily a policy violation?

The clearest policy violation occurs when a journal explicitly requires disclosure of a particular AI activity and an author fails to provide it.

For example, Elsevier's updated journal policy requires authors to declare substantive AI assistance in manuscript preparation, including relevant changes to sentence structure, organization, or content. Basic spelling, grammar, and punctuation checks are exempt from that declaration requirement. AI use in research methods must instead be documented appropriately in the Methods section.

Consequently, the same omission may have different implications depending on whether the tool corrected punctuation, drafted scientific arguments, or performed part of the analysis.

Other publishers and journals may apply different thresholds. Researchers should not assume that one publisher's exemption automatically applies to another journal.

The relationship between disclosure and the different treatment of AI grammar correction and content generation is particularly relevant when determining whether a declaration was required.

When does nondisclosure become an authorship or contribution issue?

Scholarly authorship involves more than producing sentences. It includes intellectual contribution, approval of the work, and responsibility for its accuracy and integrity.

The Committee on Publication Ethics (COPE) and the ICMJE state that AI tools cannot be listed as authors because they cannot fulfill the responsibilities associated with authorship.

However, excluding AI from the author list does not mean its contribution should always remain invisible. An author who uses AI to develop substantive interpretations or arguments may need to explain that assistance so readers and editors can understand how the manuscript was prepared.

An authorship-related concern becomes especially consequential when the author claims to have independently performed intellectual work that was substantially delegated to a system and not critically evaluated.

Even then, the issue is not that AI should receive coauthorship. It is whether the human authors have satisfied the relevant authorship criteria and represented their work accurately.

When does nondisclosure raise a research-integrity concern?

Research integrity extends beyond compliance with a submission form. It concerns the reliability, transparency, and accountability of the research process.

Suppose an AI system was used to classify interview responses, generate analytical code, or interpret experimental results. If that involvement is omitted from the methodology, readers may lack information needed to evaluate how the findings were produced.

The concern is then methodological transparency, not merely acknowledgment of a writing tool.

By contrast, undisclosed assistance with a minor grammatical correction may have little bearing on the evidential reliability of a study, particularly where the journal expressly exempts that activity from disclosure.

The significance of nondisclosure therefore depends partly on whether the omitted AI activity affected research decisions, evidence generation, analytical procedures, or scientific interpretation.

Is undisclosed AI use automatically research misconduct?

No. Formal research misconduct is defined by applicable institutional, national, funding, or regulatory frameworks. These definitions are not identical.

For example, the United States federal research-misconduct framework administered by the Office of Research Integrity focuses on fabrication, falsification, and plagiarism in proposing, performing, reviewing, or reporting research. An AI-disclosure omission does not automatically satisfy those definitions.

However, an AI-related act that involves fabricated data, falsified research records, or plagiarism may fall within a formal misconduct framework if its other requirements are established.

The ICMJE explicitly recognizes that nondisclosure of AI use may require corrective action and may be construed as misconduct in some circumstances. This does not mean every omitted declaration is automatically a proven misconduct finding.

Does intent matter when evaluating undisclosed AI use?

Intent may matter, but its significance depends on the governing rules and the type of allegation.

An author who misunderstands a recently updated disclosure requirement is in a different position from someone who knowingly provides a false declaration to conceal prohibited assistance. Both situations may require correction, but their culpability and consequences need not be identical.

For formal misconduct allegations, applicable standards may specify requirements concerning intentional, knowing, or reckless conduct. An ordinary policy breach may not require the same findings.

Intent also cannot replace examination of the consequences. An accidental omission involving a consequential analytical procedure may still need correction to preserve the reliability and transparency of the published record.

Does the extent of AI assistance determine whether disclosure is necessary?

The extent of assistance matters, but not simply as a percentage of AI-generated words.

A short AI-generated interpretation could materially affect a manuscript's scientific conclusions. Meanwhile, extensive mechanical spelling corrections might have no meaningful effect on its intellectual content.

Researchers should consider the function of the AI assistance, whether it influenced scientific substance, and the journal's specific requirements.

Some policies distinguish basic language correction from substantive rewriting. Others require broader reporting of AI use. The appropriate approach is to apply the governing policy rather than invent a numerical threshold.

This distinction is examined further through whether the extent of AI language editing changes disclosure obligations.

Can disclosure correct an earlier omission?

Sometimes. The appropriate remedy depends on when the omission is discovered, what the tool contributed, and whether the underlying activity was permitted.

Before submission, authors can revise their declarations and ensure that the Methods or acknowledgment sections accurately describe relevant AI use.

During peer review, an author may need to notify the editor and provide a corrected statement. After publication, a journal may consider a correction or another editorial response under its policies.

Disclosure cannot retroactively validate fabricated evidence or prohibited conduct. If the underlying AI-assisted work is scientifically unreliable, additional investigation or correction may be necessary.

Why is an AI-use declaration not the same as citing AI as an author?

An AI-use declaration describes a tool's role in producing or preparing research. Authorship identifies people who meet established contribution and accountability requirements.

COPE's position states that AI systems cannot assume authorship responsibilities. The ICMJE similarly places responsibility for AI-assisted material on human authors.

Accordingly, researchers should not attempt to resolve nondisclosure by adding an AI system to the author list. Instead, they should document the assistance in the location required by the journal.

Where the tool materially contributed to research procedures, an adequate Methods description may be necessary for reproducibility. Where it assisted with writing, a declaration or acknowledgment may be appropriate.

Three Different AI Omissions in One Research Project

The three omissions do not necessarily have the same significance. The distinction lies in the role of the tool, the information withheld, and the implications for scientific accountability.

This example also shows why accurately documenting AI-assisted research methods is a different obligation from merely acknowledging language assistance.

Misconceptions About Undisclosed AI Assistance

How to Assess and Correct Undisclosed AI Use

When evaluating an AI-disclosure concern, begin with the actual activity and the policy that governed the work. Avoid treating the presence of AI as sufficient evidence of misconduct or the absence of copied text as proof of compliance.

A defensible assessment distinguishes what was required, what actually occurred, and what consequence the omission had for the scholarly record.

Researchers should maintain enough documentation to explain their AI-assisted workflow without relying on memory at the point of submission. The same principle applies when establishing responsibility for errors introduced during AI-assisted editing.

What to Verify When AI Assistance Has Not Been Disclosed

Frequently Asked Questions About Undisclosed AI Use

Classify the Conduct Before Deciding Its Consequences

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.