What happened
Two university students received zero marks after being accused of using AI in their coursework. Both challenged how their universities handled the accusations, and both complaints were found justified.
The cases were reported on 15 July 2025 by the Office of the Independent Adjudicator for Higher Education (OIA), which handles student complaints against universities in England and Wales.
The OIA published six cases involving AI and academic misconduct. It raised concerns about universities failing to explain why they believed students had used AI improperly and whether students had been given a fair chance to defend their work.
What an auditable version would have shown
In the first case, an autistic student was accused of including AI-generated content in an essay.
According to the OIA, the student explained that the detection software had previously flagged another essay incorrectly. The student also provided evidence showing how they had planned and prepared their work.
The university’s panel examined the essay but did not properly consider the student’s explanation or supporting evidence. It found misconduct and gave the student a zero mark, with a capped reassessment.
The OIA found the complaint justified. When the university reconsidered the case, it found no academic misconduct.
In the second case, Turnitin flagged substantial AI-generated content in an international student’s coursework. The student was called to a viva, found to have committed misconduct and given a zero mark.
It was reported that the OIA found the complaint justified because the university had not shared any of the evidence it relied on before the viva. The student had been asked to defend their work without first seeing that evidence.
Where the gap was
The OIA recommended that the universities reconsider the cases and improve their procedures. Students needed to know what offence they were accused of, why they were being investigated and what evidence was being used. They also needed an opportunity to respond and present mitigating evidence.
Of the six cases published, two complaints were justified, one partly justified, two not justified and one settled.
In the partly justified case, the OIA found that the university had not explained what evidence supported its conclusion about AI use. It had also failed to consider whether Turnitin might be less reliable for students whose first language was not English.
In one complaint found not justified, the student had not declared their use of AI or referenced the AI-generated material. In another, a student used an AI detector to claim that a marker’s feedback was AI-generated. The OIA said these tools could be unreliable and found it reasonable for the university to give more weight to the staff member’s account.
The universities were not named.
A VerificationGate is designed to check an output against a trusted reference before it is acted on, such as treating a detection score as a prompt to review the work and the student’s own evidence.
A ConductRecord keeps what evidence was used in each case, when it was shared with the student and how the student’s response was weighed.
VerificationGate and ConductRecord are Headlights designs.
What governance should have looked like
The OIA’s casework note reported concerns about how universities investigated suspected AI use, particularly when students were not shown the evidence or given a proper chance to explain their work.
The OIA expects universities to follow its recommendations fully and on time, although its decisions are not binding on students.
When software flags a student’s work as potentially AI-generated, best practice would be to share the evidence before any hearing, consider the student’s drafts and notes, and explain the reasons for the final decision.
Failure Pattern: according to the higher education ombudsman, in two published cases universities penalised students for suspected AI use after AI-detection flags, through procedures that did not share the evidence or properly weigh the student’s own.
Governance Principle: a flag from AI-detection software should be treated as a reason to look closer, with the evidence shared and the student heard, not as proof on its own.
The reference implementation of VerificationGate and ConductRecord is open source. It lives at github.com/saffronandindia/headlights-oss, Apache 2.0 licensed and free to install. The repository is public now.
Sources
- OIA publishes a casework note and case summaries relating to AI and academic misconduct (Office of the Independent Adjudicator, 15 July 2025)
- Casework note: complaints relating to AI and academic misconduct (OIA)
- AI and academic misconduct, case summary CS072501 (OIA)
- AI and academic misconduct, case summary CS072502 (OIA)
- AI and academic misconduct, case summary CS072503 (OIA)
- AI and academic misconduct, case summary CS072504 (OIA)
- AI and academic misconduct, case summary CS072505 (OIA)
- AI and academic misconduct, case summary CS072506 (OIA)
- Alternative dispute resolution: our scheme (OIA)