03 · What You Need to Know
Responding to Unauthorized AI Use With Evidence and Procedural Fairness
Why an Explicit AI Restriction Matters
Research supervisors may establish restrictions to protect independent learning objectives, confidential information, methodological validity, or other legitimate research obligations.
When a restriction has been clearly communicated and falls within the supervisor's authority, students are expected to comply with it.
However, an explicit rule should still be examined in context.
For example, a prohibition on AI-generated thesis arguments may differ from a prohibition on uploading participant records to unauthorized external services.
The first may concern intellectual contribution and assessment validity. The second may involve confidentiality, consent, information security, and research ethics.
Different violations can require different institutional responses.
Supervisors should therefore identify the specific conduct prohibited and the obligation the restriction was intended to protect.
First, Establish What Actually Happened
A supervisor should begin with factual clarification rather than an immediate conclusion about misconduct.
Relevant questions include which AI system was used, what task it performed, when the interaction occurred, and how its outputs influenced the submitted research.
The student should have an appropriate opportunity to explain the circumstances.
For example, a student may have used AI to correct grammar in a section where the actual restriction concerned substantive argument generation.
Another may have delegated the entire analytical justification despite a clearly communicated independent-work requirement.
These activities should not be treated as equivalent without examining the wording and scope of the rule.
The supervisor should preserve relevant evidence according to institutional procedures without collecting unrelated personal information unnecessarily.
Confirm That the Restriction Was Applicable and Communicated
Before characterizing conduct as a violation, the supervisor should determine whether the restriction applied to the particular activity.
Relevant considerations include the wording of the policy, its effective date, the student's course or research context, and any permitted exceptions.
A rule introduced after the relevant work was completed should not automatically be treated as an earlier obligation.
Likewise, an informal preference should not be misrepresented as a binding institutional requirement.
Supervisors should distinguish institutional regulations from project-specific instructions and personal supervisory expectations.
Where authority or interpretation is disputed, clarification should be sought from the responsible institutional office.
Distinguish Policy Noncompliance, Research Error, and Misconduct
These categories can overlap, but they are not interchangeable.
Policy noncompliance
Conduct that conflicts with an applicable requirement, whether or not it produces an inaccurate research result.
Scientific or methodological error
A problem with research procedures, analysis, or reporting that may require correction regardless of how it originated.
Research misconduct is a separate institutional or legal classification governed by applicable definitions, evidentiary standards, and procedures.
For example, using AI contrary to an assessment restriction may constitute an academic integrity violation under university rules without necessarily satisfying a jurisdiction's formal definition of research misconduct.
Conversely, fabricating participant responses is a serious research integrity concern regardless of whether AI was used.
Supervisors should use the terminology established by the governing institution rather than assign disciplinary labels prematurely.
Why the Research Must Be Evaluated Independently of the Violation
An unauthorized AI-generated method may be scientifically correct or incorrect.
The fact that a student violated a restriction does not establish that the resulting analysis is invalid.
Similarly, a scientifically correct output does not eliminate a possible policy violation.
Supervisors should therefore examine whether the methods, evidence, and interpretations remain defensible.
Where research outputs are inaccurate, appropriate correction may be necessary regardless of the disciplinary outcome.
Where the work is scientifically sound but was produced contrary to an independent assessment requirement, reassessment may still be necessary to establish the student's competence.
Different AI Violations May Require Different Responses
| Possible violation |
Primary concern |
Potential response |
| AI assistance during a prohibited independent assessment |
Assessment validity and academic integrity |
Apply assessment and institutional procedures |
| AI-generated thesis arguments contrary to an explicit restriction |
Intellectual contribution and compliance |
Assess contribution, understanding, and applicable rules |
| Unauthorized upload of confidential research data |
Privacy, information security, and research ethics |
Follow applicable incident-reporting and containment procedures |
| AI-generated fabricated references submitted as genuine |
Evidence accuracy and possible misrepresentation |
Verify sources, correct the work, and assess relevant conduct |
| Unauthorized AI changes to an approved research procedure |
Protocol compliance and scientific validity |
Consult the responsible research oversight authority |
| Unreported material AI assistance where disclosure was required |
Transparency and compliance |
Establish the omission and follow applicable corrective or formal procedures |
These responses are illustrative. The appropriate action depends on institutional authority, the actual evidence, and the seriousness of the consequences.
When Should a Supervisor Initiate a Formal Referral?
A formal referral may be appropriate when there is credible evidence of conduct falling within the institution's academic integrity, research misconduct, or other reporting procedures.
The supervisor should identify the relevant policy and follow the established route for reporting concerns.
Not every concern requires the same procedure.
An unauthorized data disclosure may require prompt notification to information security or research ethics personnel.
An alleged violation of an independent assessment rule may be handled through academic integrity procedures.
Suspected fabrication or falsification may require referral under the institution's research misconduct framework.
Supervisors should not conduct unauthorized disciplinary investigations or promise outcomes beyond their role.
What Does Procedural Fairness Require?
Procedural fairness generally involves applying the institution's established rules, evaluating relevant evidence, and providing the student with the opportunity to respond required by those procedures.
The precise rights, evidentiary standards, and appeal mechanisms depend on the institution and jurisdiction.
Supervisors should avoid treating allegations as established findings.
They should also avoid public accusations, unnecessary disclosure of student information, or predetermined sanctions.
The European Code of Conduct for Research Integrity emphasizes responsible handling of potential violations and the importance of fair procedures.
Its principles should be applied alongside the institution's actual regulations.
Why AI-Detection Scores Are Not Sufficient Proof
Automated AI-text detectors attempt to classify the likely origin of text.
They do not directly establish whether an institutional restriction was violated.
Such tools may produce false positives and false negatives.
Liang and colleagues (2023) reported that several GPT detectors exhibited bias against non-native English writing under the conditions examined.
Even an accurate indication that AI contributed to text would not establish the precise nature of the assistance or whether it was prohibited.
Supervisors should rely on relevant evidence and applicable institutional procedures rather than treating detection scores as definitive proof.
Should Supervisors Demand Complete AI Interaction Histories?
Relevant interaction records may help establish what AI contributed to the research.
However, complete histories may include unrelated personal conversations, confidential material, and information belonging to collaborators.
Any request for records should have an appropriate institutional basis and a defined purpose.
Targeted excerpts, research decision records, or original analytical materials may sometimes provide sufficient evidence.
Where formal investigations require preservation or examination of records, authorized institutional procedures should govern access.
The question of supervisory review of AI prompts and interaction logs should therefore be considered separately from the assumption that all conversations are automatically accessible.
What If the Student Admits the Violation?
A student's acknowledgment may clarify important facts, but it does not eliminate the need to apply the appropriate procedure.
The supervisor should establish what was admitted, which rule applied, and whether additional issues remain unresolved.
For example, a student may acknowledge using AI to draft a prohibited methodology section while disputing whether the assistance influenced the analytical decisions.
The admission should be documented accurately rather than expanded into unsupported conclusions.
Any formal outcome should be determined by the authorized process.
What If the Student Claims They Misunderstood the Rule?
The supervisor should examine whether the restriction was sufficiently clear and accessible.
For example, a statement prohibiting "AI-generated content" may be ambiguous if it does not distinguish substantive drafting from grammar correction or other automated features.
A genuine misunderstanding does not automatically eliminate responsibility under an applicable rule.
However, clarity, communication, and the circumstances of the conduct may be relevant to the institution's assessment.
Supervisors should not resolve such questions through personal assumptions about what students should have understood.
Ambiguous expectations should also be clarified prospectively to reduce similar disputes.
What If the Violation Involved Confidential Participant Data?
Unauthorized processing of confidential research information requires particular attention because it may create obligations beyond academic assessment.
For example, uploading identifiable interview transcripts to an unapproved external service may conflict with consent arrangements, institutional data-security rules, or research ethics approval conditions.
The supervisor should follow the applicable incident-response process promptly.
Relevant authorities may need to determine whether containment, notification, deletion requests, or other corrective measures are required and feasible.
Supervisors should not promise that deleting a conversation necessarily eliminates information already processed by a provider.
Appropriate action depends on the system, contractual arrangements, and governing requirements.
What If the Student Used AI to Fabricate Research Evidence?
Fabricating research observations, participant responses, analytical results, or procedures raises serious research integrity concerns.
The issue is the misrepresentation of research evidence, regardless of whether AI generated the fabricated material.
The supervisor should preserve relevant records and follow the institution's procedures for suspected fabrication or falsification.
Research outputs potentially affected by the fabricated material should not be treated as reliable while the concern remains unresolved.
However, a formal misconduct finding should be made only through the authorized process.
Suspected fabrication should not be inferred merely because AI was used or because a result appears unusual.
Can Educational Remediation Be Appropriate?
Remediation may be appropriate when the institution's rules permit it and the concern involves inadequate understanding, misunderstanding, or remediable academic work.
For example, a student may be required to demonstrate independent methodological competence through a new assessment.
However, remediation does not automatically replace formal procedures where a reportable violation has occurred.
Nor does completing additional learning necessarily resolve unauthorized data processing or fabrication concerns.
Supervisors should distinguish educational support from disciplinary authority.
The appropriate response depends on the governing policy and the actual conduct.
How Should the Severity of a Violation Be Considered?
Different violations may have different consequences for assessment validity, research integrity, participant protection, and institutional obligations.
Relevant considerations may include the nature of the rule, significance of the AI contribution, actual or potential harm, and whether research outputs were affected.
However, supervisors should not invent their own sanctioning framework where institutional regulations already specify the process.
Questions of intent, repetition, and mitigating circumstances should be considered only in accordance with the applicable procedure.
A proportionate response requires both attention to the consequences and respect for institutional authority.
What If the Supervisor Previously Gave Conflicting Instructions?
A student may have received inconsistent messages from a supervisor, instructor, graduate school, or research team.
For example, one communication may permit AI-assisted drafting while another prohibits AI-generated methodological explanations.
The supervisor should examine which requirements were authoritative and applicable at the time.
Where the conflict cannot be resolved informally, the responsible academic or research governance office should provide clarification.
Students should not be expected to infer a definitive hierarchy of rules from contradictory informal statements.
Future expectations should be documented clearly, but retrospective disciplinary conclusions must follow the actual institutional framework.
How Can Supervisors Prevent Similar Violations?
Prevention begins with explicit, task-specific expectations.
Students should understand which activities are permitted, which require prior consultation, and which are prohibited.
Supervisors should explain the reasons for restrictions, particularly when they protect independent learning, confidential data, or consequential scientific decisions.
These expectations should be communicated before the relevant work begins and reviewed when research conditions change.
The distinction between permitted and prohibited AI use in research supervision should be part of ordinary supervisory discussions.
Clear rules cannot prevent every violation, but they can reduce avoidable ambiguity.
Why the Supervisor's Own Conduct Matters
Supervisors should apply appropriate AI standards to their own research and feedback practices.
For example, a supervisor who prohibits students from uploading confidential thesis material should not independently upload the same material to an unauthorized AI service for review.
Supervisory and student permissions may differ according to role and institutional authority, but the underlying confidentiality and integrity obligations remain relevant.
Consistent conduct can help students understand that restrictions serve legitimate research purposes rather than arbitrary preferences.
