Every measure that protects academic integrity processes personal information: the submitted work, similarity and detection reports and, depending on the tool, images, audio or behavioural data. The question is therefore not whether universities may check work for plagiarism and cheating, but which tools they can use, and on what terms. Generative AI makes that judgement more pressing, because some newer tools reach much further into students' lives than a classic plagiarism check. Courts, regulators and universities in the UK, the US and Australia have started to draw the lines.
Table of Contents
ToggleThree questions every integrity tool has to answer
Is there a clear legal footing? UK universities need a lawful basis under the UK GDPR, typically public task or contract. Consent rarely works in assessment: the Information Commissioner's Office notes that freely given consent is more difficult to obtain where there is an imbalance of power, "particularly for public authorities and employers". In the US, FERPA lets a vendor receive education records as a "school official" only if it remains under the institution's direct control and uses the records solely for the purposes for which they were disclosed. Under the Australian Privacy Principles, information may only be collected where reasonably necessary, and biometric data used for automated identification is sensitive information that generally requires consent.
Is it proportionate? The least intrusive tool that does the job should prevail. The clearest judicial line comes from the US: in Ogletree v. Cleveland State University (N.D. Ohio, 22 August 2022), a federal court held that a webcam scan of a student's bedroom before a remote exam was an unreasonable search under the Fourth Amendment. "Room scans go where people otherwise would not, at least not without a warrant or invitation," the court wrote. The Fourth Amendment applies only to public institutions, but the reasoning travels.
Does a human genuinely decide? The UK has written this into statute. Since February 2026, a decision counts as "based solely on automated processing" where there is "no meaningful human involvement in the taking of the decision", and significant decisions of that kind require safeguards, including the right to contest them. In the US, courts review the process instead. In Newby v. Adelphi University (N.Y. Sup. Ct., January 2026), a misconduct case began with a Turnitin AI score of 100 per cent, and the professor wrote to the student that "their system defaults you as responsible unless you contest the claim again". The court annulled the finding. It did not rule on the detector but on the process: the student's counter-evidence was not considered, he was denied an adviser, and the same official decided both the case and the appeal.
Across all three systems, the baseline duties look alike: tell students in advance what is collected and why, bind vendors by contract, set retention periods, and assess privacy risks before deployment.
Which integrity tools hold up?
Declarations and vivas
Both generate little additional data. The declaration's weakness is evidential, not legal: at King's Business School in London, 74 per cent of students did not declare their AI use despite a mandatory declaration on the coursework coversheet (Gonsalves, 2025). Vivas test understanding directly but cost staff time. The University of Sydney's "two-lane" approach uses supervised tasks such as vivas and interactive orals to assure learning, and advises using them "sparingly and strategically".
Text-matching: the classic plagiarism check
Comparing a submission with existing sources is the established plagiarism check. What matters is a clear purpose and limited retention. The leading US case on archiving student papers, A.V. v. iParadigms (4th Cir. 2009), held such archiving to be fair use under copyright law. It did not settle the privacy question. Keeping papers permanently in a database shared across institutions, or letting a vendor use them for its own purposes, needs its own justification under the UK GDPR, the Australian Privacy Principles and FERPA alike. Checks can run under a student number instead of a name, but pseudonymised data remains personal data.
AI detection
AI detectors produce probabilities, not proof, and institutions in all three countries are drawing conclusions. Vanderbilt University disabled Turnitin's AI detector in 2023: at the vendor's own claimed false-positive rate of 1 per cent, around 750 of the 75,000 papers it submitted in 2022 could have been wrongly flagged. Australian Catholic University registered nearly 6,000 alleged misconduct cases in 2024, around 90 per cent of them AI-related, according to internal documents reported by the ABC; the university calls these figures substantially overstated. It dropped the detector in March 2025.
In England and Wales, the Office of the Independent Adjudicator for Higher Education (OIA) has upheld complaints where universities did not properly weigh students' evidence. "The responsibility is on the provider to prove that the student has done what they are accused of doing," its guidance states. Detection reports should be shared with students and weighed against other evidence, including how the student actually worked.
Remote proctoring
Camera-based proctoring reaches into the home. In Ogletree, the lack of alternatives and the short notice of the scan weighed against the university. Remote proctoring therefore needs a strong justification, a genuine alternative such as an in-person sitting, and clear advance notice. Automated analysis of video or audio, and any biometric matching, deserve the strictest scrutiny.
Process documentation
Process documentation makes the path to a finished piece of work visible: drafts, versions, sources consulted, declared tool use. Australia's higher education regulator has put it on the agenda. A TEQSA-commissioned resource from June 2026 argues that institutions "need to build infrastructure capable of gathering and analysing evidence of learning processes, not merely learning products and outcomes". It also sets limits: such data provides only "partial windows" into learning, its collection must "prioritise student agency, privacy and transparency", and the methods it recommends "avoid surveillance approaches". In the UK, the OIA advises giving students the opportunity to explain how they worked, and notes that asking for notes or drafts can help.
Where students submit their own drafts at agreed points, the intrusion is low. Automatic capture of interim versions goes further, and its assessment depends on scope and technology. Keystroke logging and screen recording go furthest, because they also capture abandoned thoughts and unrelated content. To our knowledge, no court or regulator in these jurisdictions has yet ruled specifically on process documentation in assessment.
The approaches at a glance
The table offers a rough orientation, not legal advice. Whether a tool is lawful depends on the specific deployment and the law that applies to your institution.
| Approach | Intrusion | Key condition | Critical point |
|---|---|---|---|
| Declaration | low | Clear rules on permitted AI use | Limited evidential value |
| Viva or oral defence | low | Built into the assessment design | Staff time |
| Text-matching | low to medium | Clear purpose, vendor contract, retention limits | Permanent cross-institution databases |
| AI detection | low to medium | Transparency, reports shared, genuine human judgement | False positives; never sole evidence |
| Process documentation, student-submitted | low to medium | Rules set before work begins | Access and retention |
| Process documentation, automatic | medium | Data minimisation, clear legal footing | Depends on the capture method |
| Keystroke or screen recording | high | Strong justification, narrow scope | Captures irrelevant content |
| Camera proctoring | high | Necessity, alternatives, advance notice | Privacy of the home, automated and biometric analysis |
What changes next
Two rule changes point the same way. In Australia, from 10 December 2026, organisations covered by the federal Privacy Act, typically including private providers, must disclose in their privacy policies where computer programs make, or do things substantially and directly related to making, decisions that could significantly affect individuals. A tool that only recommends can still be caught. In the UK, the reformed rules on automated decisions already apply. And institutions operating in the EU should note that the EU AI Act treats AI systems that detect prohibited behaviour of students during tests as high-risk, with obligations applying from 2 December 2027.
What universities should take from this
No tool is at once minimally intrusive, conclusive and endlessly scalable. A defensible privacy strategy therefore combines the mildest effective measures, clear rules and human judgement. No automated output should justify a sanction on its own. And the legal assessment belongs before a tool is introduced, not at the moment a suspicion arises.
Documentation for your review
Mentafy combines text-matching, AI analysis and writing process documentation. For privacy officers and legal teams, we provide documentation on how our modules process data and where their limits lie, together with our data processing agreement and the information you need for a privacy impact assessment. See our tools at a glance or talk to us directly.
Selected sources
- Information Commissioner's Office: What is valid consent? (UK GDPR guidance).
- Data (Use and Access) Act 2025, section 80: UK GDPR Articles 22A to 22D.
- OIA: Casework note on complaints relating to AI and academic misconduct, July 2025, and case summary CS072501.
- FERPA regulations, 34 CFR Part 99, in particular §§ 99.31 and 99.33.
- Ogletree v. Cleveland State University, N.D. Ohio, 22 August 2022 (analysis by the Future of Privacy Forum).
- Matter of Newby v. Adelphi University, 2026 NY Slip Op 26021, Supreme Court, Nassau County, 28 January 2026.
- A.V. ex rel. Vanderhye v. iParadigms, LLC, 562 F.3d 630 (4th Cir. 2009) (US Copyright Office summary).
- Vanderbilt University: Guidance on AI detection and why we're disabling Turnitin's AI detector, 16 August 2023.
- ABC News: University wrongly accuses students of using artificial intelligence to cheat, 9 October 2025.
- University of Sydney: Embracing the future of assessment at the University of Sydney, December 2023.
- Lodge, de Barba, Ainscough et al. (2026): Assuring quality learning in a gen AI-integrated future, resource commissioned by TEQSA.
- Privacy and Other Legislation Amendment Act 2024 (Cth): automated decision-making transparency, APP 1.7 to 1.9.
- Gonsalves (2025): Addressing student non-compliance in AI use declarations, Assessment & Evaluation in Higher Education.






No comment yet, add your voice below!