Why UK Universities Are Dropping Turnitin Over Data, Not Accuracy
By Dusan Boljevic · AI/ML Engineer at TheChecker.AI
Quick answer
Newcastle and Westminster joined the University of Southampton in October 2026, ending their Turnitin contracts, bringing the total number of UK institutions walking away to at least three, with Aberdeen, Cambridge, and others reviewing their own deals. The reason isn't that Turnitin's AI detector got worse. It's a proposed change to Turnitin's end-user license agreement (EULA) that would let the company use student and faculty submissions to develop its own AI products, a provision students and university IT bodies call a non-consensual data grab, not a detection-accuracy problem. Every AI-detection story we've covered before was about whether a score could be trusted. This one is about whether a detector can be trusted with your writing at all, before it ever runs a scan.
This isn't the false-positive story again
We've written before about universities restricting how much weight an AI-detection score can carry after false positives and a lost legal challenge over process. That wave of institutional caution was about accuracy: a score flagged honest writing, a student fought back, a court sided with the student on due process grounds, and schools recalibrated how much a number alone should decide.
This is a different mechanism entirely. Newcastle University's education committee decided to end its Turnitin relationship beyond the 2026-27 academic year, citing the "potential use of colleague and student data for AI purposes from next year," according to Inside Higher Ed's October 15, 2026 reporting of a Times Higher Education investigation. The University of Westminster is retiring Turnitin at the end of its current contract in July 2027, and Southampton announced earlier in 2026 it would end its association with the company when its current deal lapses. None of these three institutions is arguing that Turnitin's AI classifier is wrong more often than it used to be. They're arguing that the terms governing what happens to submitted work changed, and that the new terms are unacceptable regardless of how the detector performs.
What actually changed in Turnitin's terms
According to the BBC's coverage of student concerns at the University of York, Turnitin proposed updates to its institutional EULA that would let the company process student submissions "to reflect current data practices" as it expands AI features inside its product. Sam Dickinson, the York Students' Union's union affairs officer, put the core objection plainly: "If your university is using Turnitin and it's using student-submitted work to train AI services, you can't meaningfully consent to that because you have to submit your work to get marked."
Turnitin pushed back on the characterization that it trains a generative model on student writing. A company spokesperson told Inside Higher Ed, "Turnitin does not train a generative AI model, or LLM. We do train a classifier model, which identifies likely AI-written text and does not produce any content to predict what is AI versus authentically human." Separately, in the BBC's reporting, Turnitin said: "We do not use customer or student work to train the AI assistant within Turnitin Clarity. We may use anonymized student submissions to improve our detection and assessment tools." The company also postponed the planned EULA change from its original July 2026 date to September 2027, saying it wanted time to work through concerns with a university-sector representation group, UCISA.
That postponement didn't resolve the dispute. Cambridge has said it will "not accept any changes to the EULA that impact on data privacy or intellectual property," per the same Inside Higher Ed report, while continuing to offer Turnitin's services to staff and students in the meantime. Southampton said in its own supplier disclosure that it "has a responsibility to protect the data, intellectual property and academic work of our students," and that while Turnitin has "postponed changes," it "has not fully clarified what future changes may be imposed." The distinction universities are drawing isn't about trusting a detection percentage. It's about trusting a vendor's contract terms around the raw material a detector runs on.
Cost is part of the story too, and that matters for reading this honestly
Not every expert frames this purely as a data-rights stand. Thomas Lancaster, a principal teaching fellow at Imperial College London, told Times Higher Education (as reported by Inside Higher Ed) that "the AI discussions are important, but ultimately the finances are just a deeper driver here." Southampton disclosed paying £87,621 for Turnitin services across one academic year; the University of East Anglia disclosed £50,892 over a similar period. Phil Newton, a neuroscience professor at the University of Swansea, argued that Turnitin's AI-detection function has become "largely redundant" for many UK institutions because most students are assessed with take-home coursework that generative AI is "very good at" producing, meaning the core value some universities originally paid for no longer holds up regardless of the EULA dispute.
Both things can be true at once: the contract terms are a genuine, independently voiced concern, and declining budgets give institutions a second, separate reason to act on that concern now rather than renew and litigate it later. That combination, not a detection-accuracy failure, is what's driving departures.
What this means if your institution uses a detector at all
If you're a student or an educator at a school still using any AI-detection vendor, the EULA question is worth asking directly, independent of how accurate that vendor's scores are. Ask what happens to submitted text after it's scanned: Is it retained? For how long? Can it be used to train or improve a model, even an "anonymized" one? Does the agreement give the vendor rights that survive after your institution stops using the product, the way Arcadia University's internal guidance flagged a "perpetual, irrevocable, royalty-free license" clause that would outlast the contract itself?
We score text for AI-generated patterns, the same category of task Turnitin's classifier performs. We don't train that scoring model on the text you submit to our demo, and we don't claim data-handling questions are irrelevant just because our product category is the same one at the center of this dispute. If anything, this is exactly the kind of governance detail every detection vendor, including us, should have to answer in plain language before a university or an individual writer hands over a draft.
The honest version of what's unresolved
Nobody disputes that AI-generated coursework is a real, growing problem. A national AACU survey found 73% of faculty say they've personally dealt with an AI-related academic-integrity issue this year. What's unresolved is which trade-off institutions are willing to make. Keep a detection tool whose contract terms make you uneasy. Drop it and lose a widely used deterrent. Or wait and see whether Turnitin's revised September 2027 terms actually address the concern. Newcastle, Westminster, and Southampton stopped waiting. Cambridge chose to keep using the product while refusing in advance to accept the disputed terms. Both are reasonable responses to the same open question, and neither settles whether AI detection itself works.
Right now this is a UK story specifically, built on EULA terms in Turnitin's institutional contracts there: Newcastle, Westminster, Southampton, Cambridge, Aberdeen, Aberystwyth, and other universities reviewing their own deals. Nobody has separately confirmed whether similar terms or similar pushback show up in other markets yet. And dropping a detector doesn't automatically mean AI-written coursework slips through unnoticed. Westminster and other schools are leaning toward assessment redesign instead: oral components, in-class writing, process evidence like drafts and revision history. Keeping a detector in place doesn't guarantee good process either. The evidence and governance around a score matter just as much as the number itself.
If your institution still uses an AI detector, or you want to check a draft before you submit it anywhere, run it through our free demo to see a sentence-level read instead of a single opaque number. For more on how any text detector actually works under the hood, see how AI detectors work, and for a deeper look at what accuracy claims do and don't guarantee, see our research-backed accuracy breakdown.
Dusan Boljevic
AI/ML Engineer at TheChecker.AI
Dusan Boljevic writes at TheChecker.AI, covering how AI-text detection works and how students, writers and teams can use it responsibly.
Related posts
A Novel Got Pulled From France's Top Literary Prize Over One Anonymous AI Scan
An anonymous account ran a bestseller through an AI detector. Other detectors disagreed. The prize pulled the book anyway.
Read more
A Chinese Court Just Wrote Evidentiary Rules for AI-Detection Reports. No US Court Has Done That Yet.
A 99.65% AI-detection score almost decided a copyright case. The court's actual rule is more useful than the number.
Read more
Can an AI Detector Actually Catch an AI-Written Resume?
Hiring managers say they can spot an AI resume. The surveys behind that claim never test accuracy.
Read more