03 · What You Need to Know
Understanding When Individual AI Use Becomes a Collective Research Responsibility
Why One Coauthor's AI Use Can Affect Everyone
Academic manuscripts are collaborative scholarly products.
Although individual authors may be responsible for particular tasks, publication normally involves shared approval of the final work and accountability under the relevant authorship standards.
The International Committee of Medical Journal Editors, for example, expects authors to meet defined criteria concerning contribution, approval, and accountability.
Those criteria are not automatically binding across every discipline or journal, but they illustrate why authors cannot treat consequential research contributions as entirely private activities.
If AI-generated code changes the analysis, the consequences may extend beyond the researcher who generated the script.
If AI invents a reference or misrepresents a finding, the resulting error may affect the credibility of the entire manuscript.
Coauthors therefore need enough information to evaluate the scientific work they are expected to approve.
Disclosure Is Not the Same as Asking Permission
Coauthors should distinguish authorization, internal disclosure, and publication disclosure.
Internal coauthor disclosure
Communicating relevant AI assistance to collaborators so they can evaluate its influence on shared research and verify consequential outputs.
Publication disclosure
Reporting AI assistance to a journal or its readers according to the publication venue's applicable instructions.
Permission is a separate question concerning whether the AI activity was authorized in the first place.
For example, a team may permit routine AI-assisted language editing without requiring an individual notification for each interaction.
However, the intended journal may have its own disclosure requirements for language assistance.
Conversely, telling coauthors about an unauthorized upload of confidential data does not retroactively authorize the processing.
What Makes AI Assistance Material?
Material assistance meaningfully influences the research's intellectual content, methods, evidence, interpretation, or reported procedures.
For example, AI may recommend changing the primary statistical model, suggest excluding observations, or generate an interpretation that becomes central to the discussion.
Such assistance can be consequential even when only a small amount of generated text appears in the manuscript.
Materiality should not be determined solely by word count, number of prompts, or time spent using AI.
A researcher may use AI repeatedly for routine grammar suggestions without materially changing the scientific argument.
Another may accept one generated recommendation that substantially changes the reported findings.
These situations should not be treated as equivalent.
Which AI Uses Generally Warrant Coauthor Communication?
| AI-assisted activity |
Internal disclosure consideration |
Reason |
| Routine grammar correction |
May be covered by standing permission |
Usually affects expression rather than substantive research |
| Preliminary search-term suggestions |
Depends on influence on the search strategy |
Search choices may affect evidence retrieval |
| Generated literature synthesis |
Material use should be communicated |
May affect evidence selection and interpretation |
| AI-generated analytical code |
Communicate consequential contributions |
Implementation can change results |
| Statistical model recommendations |
Generally warrants discussion when adopted |
May affect scientific inference |
| AI-generated qualitative themes |
Communicate material interpretive assistance |
May influence the construction of findings |
| Substantive discussion drafting |
Communicate material intellectual contributions |
May affect arguments and conclusions |
| Processing confidential shared data |
Prior authorization may be required |
Can affect obligations across the research team |
These recommendations should be interpreted alongside the team's agreements and the publication venue's requirements.
Can Routine AI Use Remain an Individual Research Practice?
Some forms of AI assistance may be sufficiently routine that separate communication to every coauthor is unnecessary.
For example, an authorized AI tool might help a researcher improve the clarity of an email or explain a general statistical concept.
Where the research team has established standing permission for such activities, individual notifications may add little value.
However, routine assistance should not become a convenient label for concealing substantive AI contributions.
The boundary should be established through shared expectations about AI use within the research team.
Specific institutional or publisher rules may still require reporting of activities that the team otherwise regards as routine.
Why Human Editing Does Not Automatically Eliminate Disclosure Obligations
A researcher may argue that AI assistance no longer matters because the generated output was extensively revised.
However, editing does not necessarily remove the influence of an earlier recommendation.
For example, AI may suggest a theoretical interpretation that becomes the organizing argument of the discussion section.
The researcher may rewrite every sentence while retaining the original conceptual direction.
Similarly, generated analytical code may be modified before execution while still influencing the final procedure.
The appropriate question is whether AI materially contributed to the research, not whether its original wording remains visible.
Why Coauthors Need to Know About AI-Generated References
AI-generated references may contain fabricated titles, incorrect bibliographic details, or genuine publications attached to unsupported claims.
When these references enter a shared manuscript, they can affect the credibility of the literature synthesis.
Coauthors should know when consequential bibliographic or interpretive material originated from AI so that appropriate verification can occur.
Checking whether a publication exists is only the first step.
The original source must also support the specific claim attributed to it.
The broader issue of identifying fabricated academic references should not be confused with verifying the evidential accuracy of genuine citations.
What If AI Generates Statistical Code?
AI-generated code can introduce errors that are difficult to recognize from the final tables.
For example, a generated script may merge datasets incorrectly, apply unsuitable exclusion criteria, or calculate a variable differently from the approved analytical plan.
The researcher who generated the code should ensure appropriate testing and communicate consequential AI involvement to collaborators responsible for evaluating the analysis.
Wilson and colleagues (2017) emphasize scientific computing practices involving testing and documentation.
These principles remain relevant regardless of whether code was written manually or generated through AI.
Coauthors should not assume that successful execution establishes scientific correctness.
Can AI Influence Research Interpretation Without Producing Final Text?
Yes.
A researcher may ask AI to explain unexpected findings or suggest alternative interpretations.
The resulting ideas may influence the final discussion even when the researcher independently writes the manuscript.
Such assistance may be material if it substantially shapes the argument or scientific conclusions.
However, merely considering a suggestion does not mean it was adopted.
Coauthors should focus on the actual influence of AI rather than assume that every exploratory interaction requires disclosure.
Consequential interpretations should be justified using the study's evidence and relevant scholarship.
Why Internal Disclosure Supports Collective Verification
Research teams often divide responsibilities according to expertise.
One collaborator may lead the statistical analysis, another may develop the literature synthesis, and another may coordinate manuscript preparation.
This division of labor does not eliminate the need for appropriate collective accountability.
If one author uses AI to perform a consequential task, relevant collaborators should know enough to evaluate the resulting work.
The issue of who verifies AI-assisted contributions produced by a coauthor should be resolved through assigned responsibilities and appropriate review.
Internal disclosure supports that process by making the contribution visible.
Does Undisclosed AI Use Automatically Constitute Research Misconduct?
No universal conclusion applies.
The significance of nondisclosure depends on the nature of the AI assistance and the requirements that applied.
For example, failing to report routine assistance that did not require disclosure may not constitute a violation.
By contrast, deliberately concealing a material contribution contrary to an explicit institutional or publisher requirement may raise a significant integrity concern.
Formal research misconduct is defined and assessed under applicable institutional or legal frameworks.
The European Code of Conduct for Research Integrity emphasizes honesty and accountability, but it should not be interpreted as establishing that every omitted AI disclosure is automatically misconduct.
Concerns should be evaluated using relevant evidence and appropriate procedures.
What If the Research Team Never Discussed AI Disclosure?
The absence of an explicit team agreement can create uncertainty about expectations.
However, it does not eliminate obligations imposed by institutional policies, research ethics approvals, or journal requirements.
Collaborators should establish what assistance occurred and determine which requirements apply.
Where material AI contributions have not been communicated, the team should address them before approving the manuscript.
New internal expectations should be communicated clearly rather than retrospectively presented as previously binding rules.
The priority is to ensure accurate reporting and adequate verification of the research.
Should Coauthors Disclose AI Use Before or After Manuscript Submission?
Material AI contributions should ordinarily be communicated before the manuscript receives final coauthor approval.
This allows collaborators to evaluate the work and determine whether additional verification or publication disclosure is necessary.
Waiting until after submission may create avoidable difficulties, particularly when the journal requires an AI-use statement.
However, the appropriate timing for different research activities may be established through the team's workflow.
For example, consequential analytical assistance may warrant discussion during analysis rather than only during manuscript preparation.
The objective is timely communication that enables meaningful review.
How Do Journal Policies Affect the Answer?
Publication venues may impose specific requirements concerning AI-assisted technologies.
The International Committee of Medical Journal Editors provides recommendations that address disclosure and human author responsibility.
Other journals and publishers may establish different rules concerning drafting, language editing, images, or analytical assistance.
Authors should consult the current instructions of the intended journal.
Internal agreement cannot override a publication venue's binding submission requirements.
The corresponding author should obtain accurate information from collaborators before making any required declaration.
Can One Coauthor's AI Use Affect Authorship?
AI assistance does not automatically determine whether a human collaborator qualifies as an author.
Authorship depends on the applicable criteria for scholarly contribution and accountability.
The International Committee of Medical Journal Editors states that AI-assisted technologies should not be listed as authors because they cannot meet human authorship responsibilities.
However, the intellectual contribution of the researcher using AI still requires evaluation.
For example, using AI to generate code does not by itself establish that the researcher made a substantial scholarly contribution.
The separate question of how AI-assisted contributions should affect authorship decisions depends on the human work performed and the relevant disciplinary criteria.
What If a Coauthor Discloses AI Use Only After a Problem Is Found?
The team should establish what AI contributed and whether the affected research requires correction.
For example, if an AI-generated literature summary misrepresented several publications, the relevant claims should be checked against original sources.
If AI-generated code affected the results, the analysis may require independent review or reanalysis.
The team should also determine whether the omission violated an applicable disclosure requirement.
Scientific correction and possible policy noncompliance are distinct questions.
The response should be evidence-based rather than assume that delayed disclosure necessarily indicates deliberate concealment.
What If a Coauthor Refuses to Explain Their AI Use?
A refusal may prevent the team from adequately verifying consequential work.
For example, if a collaborator will not explain how AI-generated analytical recommendations influenced the final model, the remaining authors may lack sufficient information to approve the findings responsibly.
The team should identify what information is necessary and whether it can be obtained through alternative research records.
Where the concern remains unresolved, manuscript approval may need to be delayed.
The issue of responding to a coauthor who refuses to disclose material AI assistance requires attention to research accountability and fair collaboration procedures.
Why Confidentiality Can Matter More Than Disclosure Timing
A collaborator may use AI to process unpublished manuscripts, confidential datasets, or participant records.
If the processing was unauthorized, telling the other authors afterward does not necessarily resolve the problem.
Research ethics and data-protection requirements may require specific action.
For example, uploading identifiable interview transcripts to an unapproved external service may create an institutional incident-reporting obligation.
The team should distinguish the failure to communicate from the underlying processing activity.
Authorization should be established before restricted information is shared.
How Can Research Teams Prevent Undisclosed Material AI Use?
Teams should agree on what counts as material assistance and which activities require internal reporting.
They should identify who verifies AI-generated references, code, and interpretations.
They should also establish how AI contributions will be documented before manuscript approval.
These expectations should be communicated early and reviewed as research tasks change.
The goal is not to monitor every routine interaction, but to ensure that consequential AI involvement does not remain invisible to collaborators who share responsibility for the resulting scholarship.
