Students describe panic, delayed investigations, and allegations built around unreliable artificial intelligence (AI) detection software. The University says Turnitin is only one piece of evidence. But on the ground, tutors, advocates, and students tell a more complicated story.

In March this year, a University of Melbourne student who has chosen to remain anonymous opened an email that would derail the beginning of their semester.

Four months earlier, they had submitted an essay during what they describe as “an incredibly difficult time,” supported by an extension and medical documentation. Now, midway through a new semester and approaching the final year of their degree, they were being formally accused of academic misconduct for allegedly using generative AI.

“The allegation was formally administered in March, four months after I had submitted the essay,” the student told Farrago. “I felt incredibly stressed, distraught and confused.”

The evidence against them was limited: inconsistent referencing and a Turnitin report claiming their essay was more than 50 per cent AI-generated.

At their Zoom hearing, they expected interrogation. Instead, the meeting lasted around five minutes.

“I was asked only one question,” they said. “Is there any reason why you think your writing might have come up as AI generated?”

Then came the statement that left them stunned.

“During the hearing, I was then told that “I know as well as they do how unreliable the detection technology is.”

The allegation was dropped the following week.

Their experience sits at the centre of  growing tension across Australian universities: as institutions race to contain the rise of generative AI, students and educators are increasingly questioning whether the systems designed to detect AI misuse are themselves reliable, fair, and safe.

At the University of Melbourne, Turnitin’s AI writing detector has become one of the most controversial tools in the academic integrity process. The University insists the software is not used as definitive proof of misconduct. Yet interviews with students, tutors, and advocacy representatives suggest that in practice, the line between “flag” and “evidence” is often blurred.

“We Do Not Believe the Tool is Reliable Enough”

The University of Melbourne formally introduced Turnitin’s AI detection technology in April 2023, shortly after the mainstream arrival of ChatGPT.

In a statement to Farrago, Deputy Vice-Chancellor (Education) Gregor Kennedy said Turnitin’s AI detector is only intended to flag assignments for further review.

“Turnitin produces an AI writing report, which alone is not a sufficient basis for an allegation of misconduct,” he said. “The University requires staff to consider additional evidence before making a report of misconduct.”

The University also said students accused of misconduct are directed toward support through the University of Melbourne Student Union (UMSU) Advocacy service.

But UMSU Advocacy says concerns about the software have existed since the beginning.

“We do not believe the Turnitin AI detection tool is reliable enough to be used on its own to raise a concern about a potential breach of academic integrity,” UMSU Advocacy  told Farrago.

According to UMSU, the months following the rollout in 2023 saw “a lot of panic,” with many allegations being allegedly raised “based solely off the Turnitin detection score.”

Although the service says staff training and policy reforms introduced in 2025 have improved the situation, it maintains that Turnitin scores should never be treated as conclusive evidence.

“It Disproportionately Hinders International Students”

Behind the scenes, tutors describe a system still struggling to define what counts as proof in the age of AI.

One University of Melbourne tutor and assignment marker, who spoke anonymously to Farrago, said early investigations relied heavily on Turnitin’s AI similarity reports.

“You would look at the Turnitin similarity report,” they said. “If it was say 20 or 30 per cent, you would bring that up with your course convenor.”

According to the tutor, Turnitin identifies patterns in sentence construction and syntax associated with large language models. But those same patterns can emerge from entirely human writing.

“Students who write relatively formulaically, particularly students who don’t have English as a first language and have been taught how to put sentences together in this highly structured manner—they’re the ones who tend to get flagged.”

The implication is troubling: the students most vulnerable to suspicion may be those writing in the careful, systematic structures they were explicitly taught to use.

“It disproportionately hinders international students or students whose English is not their first language,” the tutor said.

Ironically, they added, the inconsistencies that detectors often interpret as “human” are precisely what careful writers try to avoid.

“It’s the small little inconsistencies in writing that attest to it being written by humans.”

A System Stretched Beyond Capacity

The University’s official position is clear: Turnitin alone cannot substantiate misconduct.

But interviews suggest a widening gap between policy and practice.

Tutors say they are increasingly expected to rely on “human investigation” after confidence in Turnitin’s detector deteriorated. Yet many also say the University has not provided the resources necessary to conduct those investigations properly.

“You get paid 30 minutes per essay,” the tutor said. “To do a real, in-depth analysis of whether it’s written by AI takes you like two hours.”

As allegations multiplied, the system became overwhelmed.

“There are months-long backlogs because there is so much AI misconduct to work through,” they said. “The pressure is on us to not send them so much stuff.”

The result is an uncomfortable contradiction. Staff are discouraged from relying solely on AI detection software, whilst lacking the time and training to independently verify suspicious work.

“It was introduced as a cost-saving mechanism,” the tutor said. “You would be using an artificial intelligence program to protect against artificial intelligence during the writing process.”

“It’s a very perverse situation.”

The Students Caught in the Middle

For students, even dismissed allegations can carry enormous psychological consequences.

The student interviewed by Farrago described months of uncertainty while waiting for a resolution.

“This was an incredibly distressing experience that I was having to deal with in the middle of a brand new uni semester,” they said.

Although they praised staff for answering questions clearly and directing her toward support services, the broader process left them shaken.

“I worked on that essay across many hours,” they said. “I was incredibly offended by the allegation.”

UMSU Advocacy says such experiences are not isolated.

“Yes, there has certainly been cases where students were wrongly alleged to have improperly used AI,” the service said.

Many students are asked to provide drafts, notes, editing histories, and verbal explanations of their own work during informal meetings with Academic Integrity Officers. According to UMSU, these meetings have become especially common in AI-related cases.

The University is now pivoting toward what it calls “secure assessments,” including oral exams and supervised tasks designed to minimise AI use entirely. But for some educators, the move raises deeper questions about what universities may lose in the process.

“I still insist on the absolute centrality of the essay,” the tutor said. 

Essays, they argued, are not simply assessment tools but ways of teaching students how to think, structure arguments, and develop intellectual independence.

However, the controversy surrounding AI detection has already exposed a deeper institutional problem: trust.

Universities are trying to preserve academic integrity in an environment where technology evolves faster than policy. Students are being told to engage with AI responsibly while navigating systems that can misidentify their work. Tutors are expected to police misconduct without sufficient time or training. Advocacy groups warn against overreliance on flawed software, even as universities continue deploying it.

And underlying it all is a question nobody seems fully able to answer:

What does proof look like in the age of generative AI?

Image source: Turnitin