03 · What You Need to Know
Transparency and Accountability When Supervisors Use AI to Evaluate Student Research
Why Supervisor AI Disclosure Matters
Research supervision involves an asymmetry of academic authority.
Supervisors provide feedback that can influence research design, thesis revisions, scholarly confidence, and academic progression.
Students may reasonably assume that supervisory comments represent the supervisor's own evaluation unless another process has been explained.
When AI materially contributes to that evaluation, disclosure can help students understand how feedback was developed and what level of human review occurred.
However, transparency should not be confused with a requirement to disclose every technological operation.
The relevant issue is whether AI involvement materially affects the student's understanding of the feedback, the evaluation process, or the handling of their research material.
What Does Disclosure Mean in Supervisory Review?
Disclosure involves communicating that AI contributed to the preparation or evaluation of supervisory feedback.
It may identify the general purpose of AI assistance, the nature of the material processed, and the extent of human verification.
For example, a supervisor might explain that an authorized AI system helped identify possible inconsistencies in a draft, but all substantive comments were independently evaluated before being communicated.
This differs from providing the student with the complete prompt history.
Disclosure should be sufficiently informative to avoid misleading impressions without unnecessarily exposing confidential information.
Distinguish Disclosure From Authorization
Transparency and permission serve different purposes.
Disclosure
Informing the student that AI contributed to reviewing their research and explaining the relevant nature of that assistance.
Authorization
Establishing that the proposed AI processing is permitted under applicable institutional, confidentiality, research ethics, and other requirements.
A supervisor may disclose that AI reviewed a manuscript without having established whether the document was authorized for external processing.
Conversely, an institutionally approved AI workflow may still warrant student-facing transparency when automated feedback materially influences academic evaluation.
One requirement does not automatically satisfy the other.
Why Not Every AI-Assisted Activity Requires Identical Disclosure
AI assistance can range from routine language editing to substantive evaluation of scientific work.
For example, a supervisor may use an authorized writing tool to improve the grammar of an email accompanying feedback.
Another supervisor may use AI to recommend rejecting a student's analytical approach.
The latter activity has greater implications for the student's research and may warrant more explicit disclosure.
The appropriate threshold should reflect the nature and consequences of the assistance, as well as applicable institutional requirements.
A materiality-based approach can support meaningful transparency without making routine academic communication unnecessarily burdensome.
Which Supervisory AI Uses Most Clearly Warrant Disclosure?
| AI-assisted activity |
Disclosure consideration |
Reason |
| Grammar editing of a feedback email |
Follow applicable rules; individual disclosure may be unnecessary |
Usually affects expression rather than evaluation |
| Organizing supervisor-developed comments |
May be covered by a standing transparency statement |
Assistance may not change substantive judgments |
| Generating preliminary thesis review comments |
Disclosure is generally advisable |
AI contributes to the feedback process |
| Identifying alleged methodological errors |
Material AI involvement should be communicated |
Criticism may affect consequential revisions |
| Recommending changes to research interpretations |
Disclosure and verification are important |
AI may influence scientific conclusions |
| Evaluating thesis readiness |
Apply institutional assessment and transparency requirements |
Potentially affects academic progression |
| Processing confidential student research |
Confirm authorization before processing |
Disclosure alone does not resolve confidentiality obligations |
These are proposed supervisory practices rather than universally binding disclosure rules.
What Should Students Be Told About the AI Tool?
The appropriate level of detail depends on the purpose of disclosure and institutional requirements.
Students may reasonably benefit from knowing whether an AI system was used for language editing, structural review, or substantive methodological criticism.
Where confidentiality is relevant, the supervisor should be able to explain whether the processing occurred through an authorized environment.
However, a student-facing statement does not necessarily need to reproduce every technical setting or prompt.
The disclosure should provide enough information to clarify the role of AI and the supervisor's responsibility for the resulting feedback.
Why Students May Have Legitimate Confidentiality Concerns
Thesis drafts can contain unpublished findings, participant information, confidential institutional records, and intellectual contributions from collaborators.
Transferring such material to an external AI provider may raise concerns about access, retention, and permitted processing.
Supervisors should establish whether the processing is authorized before using the service.
Student awareness may be important, but personal agreement does not automatically override institutional restrictions or research ethics requirements.
Where necessary, the supervisor should consult the institution's data-governance or research ethics authority.
The broader question of protecting student research confidentiality during AI-assisted feedback requires attention to the actual processing arrangement.
Does Student Consent Make Every AI Review Acceptable?
No.
A student may agree to AI-assisted review without having authority to permit disclosure of confidential participant records or material belonging to collaborators.
Research ethics approvals, contractual restrictions, and institutional data-handling requirements may still apply.
Consent or agreement should therefore be distinguished from comprehensive authorization.
Supervisors should verify which permissions are required for the proposed processing.
Where authorization is absent, the appropriate response may be to use an approved alternative or avoid the processing.
Why AI Disclosure Should Include Human Accountability
A disclosure statement should not merely identify that AI was used.
It should clarify that the supervisor evaluated consequential recommendations and remains responsible for the final feedback.
For example, a statement that AI reviewed the thesis may leave the student uncertain whether the supervisor independently examined the research.
A more informative explanation distinguishes preliminary automated observations from the supervisor's verified judgments.
This is consistent with meaningful human oversight in AI-assisted research.
Human approval should reflect actual evaluation rather than a ceremonial endorsement of generated criticism.
Why AI-Generated Feedback Requires Verification Before Disclosure
Disclosing AI use does not establish that the resulting feedback is correct.
AI may produce plausible but inaccurate methodological criticism.
For example, it may recommend changing a statistical procedure without recognizing important features of the research design.
It may also claim that a cited publication does not support a statement when the original article actually does.
Supervisors should verify consequential observations before communicating them as established concerns.
Disclosure and verification are complementary responsibilities.
What Does Research on Automation Reliance Suggest?
Research on human interaction with automation highlights the importance of calibrating reliance on automated recommendations.
Lee and See (2004) discuss appropriate trust in automation, while Parasuraman and Manzey (2010) examine automation complacency and bias.
These perspectives suggest reasons to scrutinize AI-generated feedback rather than accept it because of fluent presentation.
However, they do not establish that every supervisor who uses generative AI becomes biased.
The practical implication is that substantive automated recommendations should be evaluated against independent research evidence.
Transparency can help clarify the process, but it cannot substitute for such evaluation.
Should Students Receive the Original AI-Generated Review?
Not universally.
Raw AI-generated feedback may contain inaccurate, repetitive, or irrelevant criticism.
Supervisors should not automatically forward such material as authoritative guidance.
In some circumstances, showing selected AI suggestions may support educational discussion, particularly when students are being taught to evaluate automated criticism.
However, the purpose and limitations should be explained.
Where institutional procedures require preservation or access to particular records, those requirements should be followed.
Student-facing transparency does not necessarily require disclosure of every prompt or intermediate response.
How Does AI-Assisted Feedback Affect Student Trust?
Students may interpret undisclosed substantive AI review as a departure from their expectations of scholarly supervision.
Others may welcome AI assistance if it improves the timeliness or usefulness of feedback.
These reactions are likely to depend on institutional culture, the quality of supervision, and how the technology is used.
It would be premature to claim that disclosure always increases trust or that undisclosed AI use necessarily damages it.
Nevertheless, clear communication can reduce uncertainty about the source and authority of feedback.
Trust should be supported by accurate information and accountable practice rather than assumptions about the technology.
Why Transparency Should Be Reciprocal
Supervisors may require students to disclose consequential AI assistance in thesis preparation.
Students may reasonably question why supervisors would not apply comparable transparency principles when AI materially contributes to evaluating the same work.
However, identical disclosure rules are not always appropriate because students and supervisors have different responsibilities.
The underlying principle is consistency in scholarly accountability.
Both parties should accurately represent consequential AI contributions and follow the requirements governing their roles.
Transparency should not operate exclusively in one direction without a defensible reason.
How Should Institutions Establish Disclosure Expectations?
Institutions may benefit from clarifying when AI-assisted supervisory feedback requires student notification.
Policies can distinguish routine editorial assistance from substantive automated evaluation.
They should also address authorized processing environments, confidentiality, record handling, and responsibility for final judgments.
Requirements should be communicated before substantial AI-assisted review becomes part of supervision.
Such arrangements can complement clear expectations for student AI use while recognizing the distinct responsibilities of supervisors.
How Should Disclosure Work in Collaborative Supervision?
A thesis may be reviewed by several supervisors, committee members, or external advisers.
Different reviewers may use different AI tools or operate under different institutional requirements.
Where AI materially influences substantive feedback, the relevant participants should understand the applicable transparency and confidentiality expectations.
For example, one institution may authorize a particular AI processing arrangement while another restricts it.
Reviewers should not assume that permission granted in one setting automatically applies to shared research material.
Appropriate coordination can reduce uncertainty about who processed the document and who remains responsible for the evaluation.
Should Disclosure Differ for Formative Feedback and Formal Assessment?
Formative feedback is intended primarily to support learning and improvement.
Formal assessment may determine thesis progression, grades, or degree completion.
Both require accurate and responsible evaluation, but formal decisions may involve additional procedural requirements.
AI-assisted formative review may be appropriate when authorized and verified.
For consequential assessment, institutions should specify whether and how AI may contribute to the evaluation process.
Final academic decisions should remain with authorized human evaluators.
Students should not be misled about the role automated systems played in consequential judgments.
What If AI Feedback Is Later Found to Be Incorrect?
The supervisor should correct the affected feedback and reconsider any recommendations based on the error.
For example, if an AI system incorrectly identifies a methodological flaw, the student should not be required to make an inappropriate change merely because the criticism appeared in supervisory comments.
Where the error influenced a formal academic decision, the institution's review or correction procedures may be relevant.
Disclosure of AI assistance does not remove the supervisor's responsibility for the consequences of unverified criticism.
Accurate correction is part of accountable academic supervision.
How Can Supervisors Communicate AI Use Without Creating Unnecessary Alarm?
Disclosure should be factual and proportionate.
A supervisor can explain that an authorized AI system helped identify possible organizational inconsistencies and that the final comments were independently reviewed.
There is no need to portray ordinary authorized assistance as inherently suspicious.
Equally, the supervisor should not minimize substantive automated evaluation by describing it merely as grammar correction.
The explanation should accurately reflect what occurred.
Clear communication allows students to evaluate feedback with an appropriate understanding of its origins and limitations.
