03 · What You Need to Know
Managing AI Disclosure Refusals Without Presuming Misconduct
Why a Coauthor's Refusal Creates a Research Accountability Problem
Collaborative research depends on the ability of authors to evaluate the work they collectively approve.
Individual collaborators may perform specialized tasks, but their contributions become part of a shared scholarly product.
When one author uses AI to generate analytical code, develop interpretations, or produce substantive evidence summaries, other authors may need information about how those outputs were evaluated.
A refusal to provide relevant information can prevent the team from establishing whether the work is scientifically defensible.
The concern is not simply that AI was involved. It is that consequential research procedures may remain insufficiently documented or verified.
The team should therefore focus on the information needed to support accountable publication rather than on obtaining unrestricted access to another researcher's AI account.
First, Determine What the Coauthor Is Actually Refusing
A refusal to disclose AI use can describe several different situations.
A researcher may decline to share complete conversation histories while remaining willing to explain the analytical workflow.
Another may acknowledge AI assistance but refuse to identify which research procedures were affected.
A third may decline to provide code, original data-processing records, or other evidence needed to verify the reported findings.
These situations should not be treated as equivalent.
The team should clarify whether the disagreement concerns disclosure of AI involvement, access to private interactions, preservation of research records, or the ability to verify scientific claims.
Distinguish AI Disclosure From Access to Complete Prompt Histories
Material AI-use disclosure
Information about which consequential research tasks involved AI, what the system contributed, and how the resulting work was evaluated.
Complete interaction-history access
Access to recorded prompts, responses, uploads, and potentially unrelated conversations within an AI system.
A research team may legitimately require information about consequential analytical procedures without necessarily requiring every personal AI interaction.
Complete histories can contain unrelated conversations, confidential material, and information belonging to other projects.
The appropriate request should be determined by the purpose of verification and the requirements that apply.
What Information Should the Team Request?
The team should begin with the minimum information necessary to understand and evaluate the contribution.
For example, if AI generated statistical code, the relevant evidence may include the final script, the analytical purpose, the verification performed, and the resulting outputs.
If AI suggested references, the team may need the original publications and evidence supporting the claims.
If AI influenced qualitative interpretation, the team may need an explanation of how generated suggestions were considered within the selected methodology.
| AI-assisted activity |
Information to request |
Reason |
| Literature synthesis |
Original sources and verification of substantive claims |
Establish evidential accuracy |
| Statistical programming |
Final code, relevant tests, and analytical records |
Verify computational implementation |
| Model selection |
Methodological justification and relevant alternatives |
Evaluate analytical suitability |
| Data cleaning |
Transformation rules, exclusion records, and reproducible scripts |
Protect data integrity |
| Qualitative interpretation |
Analytical rationale and evidence supporting final interpretations |
Establish methodological coherence |
| Substantive manuscript drafting |
Description of material AI contributions and source verification |
Support intellectual accountability |
| Confidential data processing |
Authorized processing arrangement and relevant compliance records |
Establish whether information was handled appropriately |
These requests should be adapted to the actual research and any applicable institutional requirements.
Why the Coauthor's Confidence Is Not Verification
A collaborator may sincerely believe that AI-assisted work is correct.
However, confidence does not establish that the procedures were appropriate or the outputs accurate.
For example, generated code may run successfully while incorrectly merging datasets or excluding observations.
Likewise, an AI-generated literature synthesis may contain genuine references attached to unsupported claims.
Research teams should examine the relevant evidence rather than accept assurances alone.
The responsibility for verifying AI-assisted contributions produced by one coauthor should be explicit and supported by appropriate records.
Does Every Coauthor Need to Know the Same Details?
Not necessarily.
Research collaboration involves specialized expertise.
A statistical analyst may examine code and model assumptions in greater detail than collaborators responsible for theoretical interpretation.
However, all authors should have sufficient information to understand consequential research decisions and the verification arrangements supporting them.
The distinction between specialist review and collective accountability matters.
The broader expectation of coauthor awareness of material AI assistance does not require identical technical knowledge among every author.
What If the Coauthor Has Legitimate Privacy Concerns?
A researcher may reasonably object to sharing an entire AI account history that contains unrelated personal or professional information.
The team should examine whether the necessary research evidence can be provided through targeted excerpts, verified scripts, or structured decision records.
Where complete records are required by an approved protocol or authorized investigation, the applicable access and confidentiality procedures should govern their handling.
Privacy concerns do not automatically eliminate legitimate research-record obligations.
Equally, research accountability does not automatically justify unrestricted inspection of unrelated conversations.
A proportionate request should identify the information needed and the basis for requesting it.
What If the Coauthor Claims the AI Conversations Were Deleted?
The team should determine which records remain available and which documentation requirements applied when the work was performed.
Some AI systems do not preserve complete interaction histories, while others permit users to delete conversations.
Missing prompts should not be reconstructed from memory and presented as verbatim records.
However, the absence of prompt histories does not necessarily make verification impossible.
Final scripts, original datasets, analytical outputs, methodological justifications, and source documents may provide more directly relevant evidence.
If essential research procedures cannot be reconstructed, the team may need to repeat or independently verify the affected work.
What If the Refusal Concerns Confidential Research Data?
Confidentiality can legitimately restrict how research records are shared.
For example, a collaborator may be prohibited from distributing identifiable participant records outside an authorized research environment.
The solution is not necessarily to waive verification.
Instead, the team should establish an appropriate review arrangement, such as access by an authorized analyst within a secure environment.
Research ethics approvals, data-sharing agreements, and institutional policies may determine what is permissible.
The team should not require unauthorized disclosure merely to investigate possible unauthorized AI use.
How Should the Corresponding Author Respond?
The corresponding author commonly coordinates manuscript approval and communication with the journal.
However, that role does not automatically confer authority to compel unrestricted access to a collaborator's personal AI interactions.
The corresponding author should identify what information is necessary for accurate publication declarations and responsible approval.
If material AI involvement cannot be established, the corresponding author should avoid submitting an unsupported statement that the manuscript satisfies all relevant requirements.
Where necessary, submission should be postponed while the team seeks clarification or appropriate institutional guidance.
The corresponding author's coordinating role does not eliminate the responsibilities of the other authors.
Why Journal Requirements Matter
Journals and publishers may require authors to disclose specified uses of AI-assisted technologies.
The International Committee of Medical Journal Editors (ICMJE) provides recommendations concerning AI-assisted technologies and human author accountability.
Other publication venues may establish different requirements.
The team should examine the current instructions of the intended journal rather than assume one universal disclosure standard.
If a collaborator refuses to provide information necessary for a required declaration, the corresponding author may be unable to complete the submission accurately.
Internal disagreements do not justify submitting misleading declarations.
Can the Team Independently Verify the Work Without the Coauthor's Cooperation?
Sometimes.
If the final analytical code, data, and methodological records are available, another qualified researcher may be able to inspect or reproduce the analysis.
For literature claims, collaborators may independently examine the original publications.
However, independent verification of the final work does not necessarily resolve every disclosure or authorization obligation.
For example, a verified statistical result may still have been produced through unauthorized processing of confidential data.
The team should distinguish scientific verification from compliance with applicable AI-use requirements.
What If Independent Verification Reveals an Error?
The researchers should determine which outputs were affected and correct the work.
For example, an incorrect AI-generated data transformation may require rebuilding the analytical dataset and rerunning statistical models.
Changes should be documented accurately.
If the manuscript has already been submitted or published, the journal's correction procedures may become relevant.
Scientific errors should be addressed regardless of whether the original collaborator disclosed AI assistance.
However, the existence of an error does not automatically establish intentional misconduct.
Should the Team Remove the Coauthor?
Authorship should not be withdrawn automatically because a collaborator refuses a particular disclosure request.
The team must distinguish a disagreement about access to records from failure to satisfy applicable authorship criteria or institutional obligations.
Authorship changes can have significant academic consequences and may require agreement among authors, institutional review, or journal procedures.
The ICMJE provides recommendations concerning authorship responsibilities and disputes, although applicability depends on the publication context.
Unilateral removal should not be used as an informal sanction or as a way to avoid resolving legitimate research concerns.
Where an authorship dispute persists, appropriate institutional guidance should be sought.
Does Refusal Automatically Establish Research Misconduct?
No.
A refusal may arise from legitimate privacy concerns, uncertainty about requirements, disagreement over authority, or unwillingness to provide necessary research records.
These circumstances require different evaluations.
Formal misconduct findings depend on applicable definitions, evidence, and institutional procedures.
The European Code of Conduct for Research Integrity emphasizes accountability and fair handling of potential violations.
Researchers should avoid treating noncooperation as conclusive evidence of fabrication, falsification, or another specific form of misconduct.
Nevertheless, unresolved noncooperation may prevent the team from responsibly approving the research.
When Should Institutional Guidance Be Requested?
Institutional consultation may be appropriate when the team cannot resolve the disagreement or when the concern involves formal research obligations.
Relevant offices may include the research integrity office, institutional research governance office, department leadership, or another designated authority.
If the dispute concerns confidential participant information, research ethics or data-protection personnel may need to be involved.
If it concerns authorship, the institution's authorship dispute procedures may be relevant.
The appropriate route depends on the issue and governing policies.
Teams should avoid assuming that every disagreement requires a formal misconduct allegation.
How Should the Team Document the Disagreement?
Documentation should accurately distinguish requests, responses, available evidence, and unresolved questions.
For example, the team may record that the analyst provided final statistical outputs but did not provide the underlying code needed for verification.
It should not characterize that fact as proof of deliberate concealment without additional evidence.
Records should be stored through authorized arrangements and limited to relevant information.
Accurate documentation can support later clarification while reducing the risk that informal assumptions become treated as established facts.
What If the Manuscript Is Almost Ready for Submission?
Submission deadlines can create pressure to proceed despite unresolved concerns.
However, urgency does not establish that a consequential AI-assisted contribution has been verified.
If the missing information prevents responsible approval or accurate declarations, submission should generally be delayed until the issue is resolved.
The team may seek an extension, obtain qualified independent verification, or consult the relevant institution or journal.
The appropriate response depends on the significance of the uncertainty and applicable requirements.
A deadline should not become a substitute for scientific evidence.
How Can Research Teams Prevent Disclosure Refusals?
Clear expectations established before consequential AI use can reduce later disputes.
Teams should define material assistance, identify required documentation, assign verification responsibilities, and establish how concerns will be resolved.
These expectations should distinguish ordinary tool use from decisions affecting shared methods or findings.
A written research team agreement on AI use may provide a practical foundation.
However, agreements should remain consistent with institutional authority and should not demand unnecessary access to personal interactions.
Why the Final Decision Must Remain Evidence-Based
The team should avoid two opposing mistakes.
One is assuming that the coauthor's refusal proves misconduct.
The other is assuming that the refusal can be ignored because the final manuscript appears credible.
The defensible approach is to determine what evidence is necessary, whether it can be obtained, and what uncertainty remains.
Where material research claims cannot be verified, the team should not represent them as adequately established.
Where disclosure requirements cannot be satisfied, the manuscript should not be submitted with inaccurate declarations.
The objective is collective author accountability supported by fair procedures and verifiable evidence.
