03 · What You Need to Know
How to Evaluate Intent When AI Disclosure Policies Change
First Establish Which Policy Actually Applied
An assessment of nondisclosure should begin with the applicable requirement rather than the researcher's presumed intention.
AI policies may change during manuscript preparation, submission, peer review, or production. A publisher may introduce a dedicated declaration, revise the definition of reportable AI assistance, or impose additional restrictions on particular activities.
The relevant evidence includes the journal's author instructions, submission-system questions, editorial correspondence, and any institutional or funder requirements applicable at the time.
Where possible, obtain a dated copy of the policy or an archived version of the relevant instructions. A current webpage does not necessarily establish what authors were required to disclose several years earlier.
It is also necessary to distinguish policy publication from policy applicability. A rule may specify an effective date, a transition arrangement, or particular categories of submissions.
Without establishing the governing requirement, a conclusion that the author knowingly violated it would be premature.
Policy Awareness and Policy Understanding Are Not Identical
A researcher may have encountered a disclosure policy without correctly understanding its scope.
For example, an author might know that a journal prohibits listing AI systems as authors but mistakenly assume that this is the journal's only AI-related requirement. Another might interpret "AI-generated content" as excluding substantial rewriting of author-drafted paragraphs.
Such interpretations may be incorrect, but incorrect interpretation is not automatically evidence of intentional concealment.
A careful assessment asks what information was reasonably available, how clearly the requirement was communicated, and whether the author's explanation is consistent with the surrounding records.
Professional responsibility still matters. Researchers are generally expected to review applicable submission requirements. Failure to do so may constitute poor practice or noncompliance even when deliberate deception cannot be established.
What Evidence Supports an Honest Misunderstanding?
An explanation of misunderstanding becomes more credible when supported by contemporaneous evidence rather than retrospective assertion alone.
Relevant circumstances may include an ambiguous policy, documented reliance on an earlier version, correspondence seeking clarification, or records showing that the researcher openly discussed AI use elsewhere in the project.
For example, a researcher who described AI-assisted editing in internal manuscript notes but omitted a newly required declaration may have misunderstood the reporting procedure. That possibility should be evaluated alongside the actual policy and submission record.
Prompt acknowledgment and cooperation after discovery may also be relevant. However, cooperation does not prove that the original omission was innocent, just as delayed disclosure does not necessarily prove deliberate concealment.
Evidence should be assessed collectively rather than treating any single behavior as decisive.
What Evidence May Support Deliberate Concealment?
Deliberate concealment generally requires stronger evidence than the mere absence of a declaration.
Potentially relevant evidence includes an explicit submission question, a documented acknowledgment that disclosure was required, instructions to collaborators not to report AI use, or a knowingly false statement about whether AI contributed to the manuscript.
Suppose an author receives a message from the corresponding author explaining that the journal requires disclosure of AI-assisted drafting. The author replies that they understand the requirement, but subsequently insists that the submission declare no AI use despite records showing substantial AI-generated drafting.
That sequence may support an inference of intentional misrepresentation. Nevertheless, the researcher should have an opportunity to explain the circumstances, and any formal determination must follow applicable procedures.
Investigators should distinguish documented knowledge from assumptions based on seniority, disciplinary expertise, or general familiarity with AI.
Why a Timeline Is Essential
A chronological reconstruction can reveal whether the researcher encountered the requirement before or after the relevant decision.
| Event |
Evidence to examine |
Why it matters |
| AI tool was used |
Interaction logs, drafts, code histories, correspondence |
Establishes the actual contribution and approximate timing |
| Disclosure policy was introduced or revised |
Dated policy, archived instructions, publisher announcement |
Establishes the governing requirement |
| Authors prepared submission materials |
Draft declarations, internal messages, submission checklist |
Shows how the requirement was interpreted |
| Submission declaration was completed |
Submission-system records and author confirmations |
Identifies the representation made to the journal |
| Journal requested clarification |
Editorial correspondence and author responses |
May establish notice of the requirement and subsequent conduct |
| Omission was discovered |
Discovery records, author communications, correction requests |
Shows how the issue was addressed after becoming known |
The sequence matters because knowledge acquired after submission cannot automatically establish what the researcher understood beforehand.
Similarly, an explicit warning received during peer review may change the assessment of a later inaccurate declaration, even if the original omission arose from misunderstanding.
How Much Does the Clarity of the Policy Matter?
Policy wording affects what researchers could reasonably understand.
A statement requiring disclosure of "AI-generated manuscript content" may raise interpretive questions about spelling correction, sentence restructuring, translation, or automated formatting. More specific instructions can reduce ambiguity.
ICMJE guidance, for example, asks authors to describe whether and how AI-assisted technologies contributed to submitted work. It identifies writing assistance and methodological applications as activities to report in appropriate manuscript sections.
However, ICMJE recommendations do not automatically govern every journal. The actual instructions adopted by the publication remain relevant.
Ambiguity does not eliminate an author's responsibility to seek clarification where necessary, but it may affect whether deliberate violation can reasonably be inferred.
Can Negligence Be Distinguished From Honest Error and Deliberate Concealment?
Yes, although the boundaries depend on the applicable standards.
Honest misunderstanding
The researcher genuinely misinterpreted or was unaware of a requirement, without evidence that the omission was intended to mislead.
Negligent noncompliance
The researcher failed to exercise the care reasonably expected in checking or following applicable requirements, without necessarily intending deception.
Deliberate concealment
The researcher knowingly withheld or misrepresented reportable AI involvement to avoid disclosure or create a misleading account.
These distinctions are analytically useful but should not be mistaken for a universal disciplinary classification.
Under the U.S. Public Health Service research misconduct framework, fabrication, falsification, or plagiarism must be established according to specified requirements, including the relevant mental-state and evidentiary standards. Honest error and differences of opinion are excluded from its definition.
Other institutions may separately address negligent research practices, publication-policy breaches, or failures of professional responsibility even when formal research misconduct is not established.
Does the Researcher's Response After Discovery Establish Their Original Intent?
Post-discovery conduct may provide relevant evidence, but it must be interpreted cautiously.
An author who promptly provides records, acknowledges the omission, and requests correction may demonstrate a willingness to remedy the problem. That response does not conclusively establish what the author intended at submission.
Conversely, an author who initially disputes an allegation may have legitimate reasons for doing so, including disagreement about the applicable policy or the accuracy of the evidence.
Deliberate destruction of relevant records, knowingly false explanations, or attempts to prevent a necessary correction may raise additional concerns. Such behavior should be evaluated through established procedures rather than treated as automatic proof of the original allegation.
The purpose is to reconstruct the evidence fairly, not to reward a particular style of response.
Why AI Detection Scores Cannot Establish Concealment
AI-text detection tools attempt to classify text according to statistical characteristics associated with machine-generated writing. Their outputs do not directly reveal the author's decisions, policy knowledge, or intent.
A high detection score cannot establish which AI system was used, whether its output entered the final manuscript, or whether the author knowingly violated a disclosure requirement.
Human-authored academic prose may also be incorrectly classified, particularly when writing conventions are highly standardized or text has undergone substantial editing.
Detection results should therefore not be treated as a substitute for contemporaneous records, author explanations, or independent investigation.
Where an allegation rests primarily on a detector score, the evidentiary basis for claiming deliberate concealment is weak.
What Should Happen When Intent Remains Uncertain?
Not every investigation will establish a definitive explanation for the omission.
If the evidence supports a disclosure violation but does not establish deliberate concealment, the journal may still require an accurate declaration or correction. The publication record can be amended without making an unsupported finding about the author's state of mind.
Where formal misconduct proceedings are initiated, the responsible authority should apply its governing evidentiary standard and procedural safeguards.
Uncertainty should be documented rather than concealed through categorical language. A conclusion such as "the required disclosure was omitted, but the available evidence does not establish deliberate concealment" may be more defensible than either accusation or exoneration unsupported by the record.
Correcting the scholarly record and determining individual culpability are distinct responsibilities.
