TL;DR
The University of Southampton will not renew its Turnitin contract once 2026-27 finishes, and is winding down use of the plagiarism and AI-detection service immediately. It pointed to proposed licence changes that would have permitted student submissions to be used in training AI services. Turnitin denies training its own writing tool on student work.
A university using the leverage it has
Southampton said it understood the proposed changes had been shelved, but that the California company “still intends to revisit implementing these updates next year” — which is why it is not waiting for the contract to lapse quietly. Alternatives for maintaining academic integrity are being evaluated.
That is a rarer move than it sounds. Universities have issued a great many policy statements about AI and consent over the past two years; terminating a contract with a supplier used by thousands of institutions worldwide is a different order of action, and one only a large customer can make stick.
What the terms actually say
Turnitin’s position is narrower than a flat denial. It says it does not train Clarity — its in-house composition tool, which shows markers how and when a piece of work took shape — on “customer or student work”. Separately, it says it “may use anonymised student submissions” to improve detection and assessment. Those are different claims about different products.
The licence agreement is where the asymmetry sits. Students submitting from outside the EU grant Turnitin a licence described as “non-exclusive, royalty-free, perpetual, worldwide, irrevocable”, bounded by the stated purpose of providing and improving the service. Submissions from inside the EU fall under GDPR, which confines processing to what is necessary for the task. British students, post-Brexit, sit on the wrong side of that line by geography rather than by any choice they made.
Looking forward
Thomas Lancaster, who teaches at Imperial College London, said he had no inside knowledge but believed the company was handling permissions properly — while adding that universities and students have not recognised what that body of essays and assignments is worth. He expects the market to fragment, noting Anthropic’s recent move to watermark generated text as an alternative detection route, and floated a consortium of universities building their own tool with control of the underlying data. The obvious weakness of going in-house alone, he said, is that it cannot catch work copied from a friend at another institution.