90 incidents on record · 2026 Headlights Incident reports by Ellie Harris · Melbourne
10 new this week Library last updated 13 July 2026
← The incident library
HD-INC-062
Education · Australia · 2025 · Automated-decision harm

An Australian university accused thousands of students of AI misconduct using a detector it knew was unreliable

By Ellie Harris · Filed 2024 academic year; AI indicator withdrawn March 2025

Alleged: Australian Catholic University (ACU); Turnitin, LLC developed or deployed the AI system implicated in this incident. Details are drawn from public reports; parties are presumed innocent of any wrongdoing not established by an official finding.

An Australian university accused thousands of students of AI misconduct using a detector it knew was unreliable

What happened

The ABC reported that in 2024 Australian Catholic University recorded nearly 6,000 academic-misconduct referrals across its campuses, and that about 90 percent related to suspected AI use, a figure ACU said was substantially overstated. ACU had adopted Turnitin’s AI-writing indicator in 2023; the tool highlights suspected AI-generated text in blue and returns a percentage estimate of how much of a submission it judges to be machine-written. According to that reporting, internal documentation indicated the university had regularly relied on the indicator, at times as the main basis for a referral. Turnitin’s own guidance cautions that its reports “may not always be accurate” and should not be the sole basis for sanctions, and independent evaluations have found AI detectors unreliable under realistic conditions, with no tool reliably accurate across tasks. ACU said that any case in which Turnitin was the sole evidence was dismissed.

Students reportedly described months in limbo. One final-year nursing student, Madeleine, reportedly had her results marked “withheld” for six months after an essay was flagged, which she says contributed to her not being offered a graduate position she needed to register. A paramedic student told the ABC that almost his entire essay “was lit up in blue, 84 per cent of it supposedly written by AI.” Some students were reportedly asked to produce search histories and drafts to prove their innocence, reversing the burden of proof onto the accused. Internal records reportedly show ACU was aware of the tool’s unreliability for more than a year before withdrawing it in March 2025, and on review roughly a quarter of the referrals were dismissed.

What an auditable version would have shown

A probabilistic score is a lead, not a finding. An auditable version records, for each allegation, what the detector actually returned, what corroborating evidence existed beyond it, and who decided to proceed on what basis, signed when the decision was made. In aggregate it would have surfaced, early, how often referrals rested on the score alone and the false-positive exposure that implied, while that was still a warning rather than a controversy touching thousands of students.

Where the gap was

The gap was treating a model’s estimate as verification. A detector reports that a text “looks AI-generated”; it does not establish that it is, and the vendor says as much. A VerificationGate is the control: a flag is a proposal to be checked against real evidence (process history, drafts, an oral defence), never an adverse finding on the model’s say-so, and it routes the question to a trusted source rather than back to another model. A MetricRecord over signed records would have exposed how often the indicator was the sole basis, and the dismissal rate, while they still mattered. A ConductRecord ties each case to the evidence it actually stood on, so a referral resting only on the detector is visible in the file, not discovered on appeal.

What governance should have looked like

An accusation that can withhold a degree needs more than a number from a tool whose own vendor says not to rely on it alone. Corroboration required before an allegation is raised, the burden of proof kept where it belongs, and the reliability of the instrument monitored continuously against real outcomes: those are the ordinary safeguards of any high-stakes decision. Used as one signal among several, an AI indicator is a prompt to look closer; used as the finding, it turns a probability into a verdict, and that is how thousands of students were drawn into a misconduct process on the strength of a score.

The reference implementation of VerificationGate, MetricRecord, 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

The mailing list

Fresh incident reports every week. One email to match.

We add new incidents to the library regularly, and send a single short email each week with what's new. The library stays free and open; this is just how you keep up with it.

No tracking. Unsubscribe in one click.

The record

An auditable system would have produced a signed, tamper-evident record the moment this happened: what the system did, the version that did it, the basis it acted on, and the action taken, and Australian Catholic University (ACU); Turnitin, LLC could have produced it on demand.

This is the record the system as deployed did not produce in a signed, auditable form.

What this teaches
Capture what happened when it happens
What the system did, the version that did it, the basis it acted on, and the action taken, recorded at the moment, not reconstructed after.
Sign it, so no one has to trust the record-keeper
A tamper-evident entry. Edit it later and the signature breaks. The record does not ask for the benefit of the doubt.
Make it verifiable by anyone
A court, a regulator, a customer's lawyer can check the record themselves, without taking the company, or us, at our word.

Headlights summarises publicly reported AI incidents. All summaries are independently written, attributed to their original sources, and intended for research and educational purposes. Allegations are identified as such until established through official findings.

Last reviewed June 2026. This report is based on the sources listed above and reflects information available at the time of review; later developments may not be captured. Where a person is described as charged with or alleged to have done something, that allegation is unproven unless a conviction or a court or regulatory finding is stated. Headlights publishes journalism and commentary, not legal advice.

Want to write back?

Direct to my inbox.

ellie@useheadlights.com →