WHO MUST PROVE THAT A STUDENT MISUSED AI?
Suspicion may begin an academic-integrity inquiry. It should not quietly transfer the burden of proving innocence to the student.
In today’s governance note, I want you to consider what happens when an institution suspects that a student has used AI improperly in assessed work. A detection score may appear unusually high. The writing may differ from earlier submissions and, a reference cannot be verified. This may lead to an assessor who may have feelings that the work does not sound like the student.
What happens next, and who must establish what actually occurred? This question should be answered before the first allegation. If institutions wait until an individual case arises, uncertainty about the technology can become uncertainty about the student. A screening signal may gradually be treated as proof, while the student is expected to demonstrate that they wrote the work themselves. That reverses the proper order of an investigation.
I would begin by separating suspicion, evidence and finding.
An AI-detection result may provide a reason to examine work more closely. It does not explain which tool was used, how it was used or whether that use breached the rules applying to the assessment. Writing style, vocabulary and document history may add context, but each has limitations.
Institutions should define the evidentiary role of each signal in advance:
The institution should first establish what use was permitted. “AI was involved” is not the same as “academic misconduct occurred.” A student may have used spelling support, translation, accessibility software, feedback tools or generative assistance that was allowed but poorly explained.
Rules must therefore be specific to the assessment. Tell students what tools they may use, which forms of assistance require acknowledgement, what records they should retain and what will happen if concerns arise. This should be communicated before submission, not introduced during an investigation.
The student must then receive the substance of the concern and the evidence being considered. They should have a reasonable opportunity to respond, introduce relevant material and correct factual assumptions. That process must also be accessible.
A live oral defence may help establish whether a student understands their work, but it can disadvantage someone who is deaf, anxious, neurodivergent, communicating in an additional language or affected by another disability. The institution should preserve the purpose of the inquiry while offering appropriate communication formats, preparation time and reasonable adjustments.
Do not turn one narrow model of authorship into the test. Not every student produces extensive drafts. Some dictate, edit non-linearly, use assistive tools or develop an argument away from the document before writing it quickly. Absence of a conventional version history is not proof that the work is not theirs.
Current Joint Council for Qualifications guidance requires centres to maintain internal appeal procedures covering decisions to reject work for malpractice or inability to authenticate it. Its malpractice framework emphasises full investigation, fairness, impartiality and the opportunity for the person accused to respond to the evidence. JCQ also suggests that an oral discussion may assist authentication when plagiarism is suspected. JCQ non-examination assessment guidance, JCQ plagiarism guidance
Jisc’s student research found continuing uncertainty about what AI use institutions permit, despite the widespread introduction of guidance. Ambiguous rules followed by high-consequence enforcement create avoidable unfairness. Jisc student perceptions of AI. Protecting academic integrity does not require institutions to ignore reasonable concerns, but that they investigate those concerns without allowing uncertain technology to become a presumption of guilt.