A supplier shouldn’t be able to rewrite who owns students’ essays

Back in July, Turnitin was due to change its end user licence agreement so that student work could be used for “improving or developing AI tools.”

The sector intervened, and the company pulled back the day before implementation.

It first paused for 60 days after what Sussex described as coordinated intervention from sector bodies, then confirmed there would be no changes during the 2026–27 academic year.

Stirling says it has been told nothing will change before 1 September 2027. A statement circulated to universities has now confirmed that the new EULA has been postponed until September 2027 and that Turnitin will redraft it in partnership with UCISA. The changes have been stayed and are being rewritten – but they haven’t been dropped.

Despite this, the University of Southampton has still chosen to walk away after the 2026–27 academic year. Explaining the decision, it said:

We have a responsibility to protect the data, intellectual property and academic work of both our students and the University.

Sussex has moved away from Turnitin, UEA has indicated that it will, and in private conversations I’ve had with SU colleagues, other universities are signalling that a Southampton-style exit is on the cards.

Speaking with colleagues at Southampton, it’s clear that they have simply opted to direct departments to stop using Turnitin and use built in-alternatives like Blackboard’s SafeAssign – rather than buying out a costly contract.

Not what we signed up to

This is about consent to use students’ work for commercial purposes. Students can’t meaningfully consent to their work being processed to train AI or develop services when Turnitin’s processing is a mandatory part of progressing through university.

At York, university policy encourages Turnitin rather than requiring it, but at department level it can be compulsory – Health Sciences has required summative submission through Turnitin since 2023–24.

Students’ work should belong to them. As NUS Scotland president Justine Pedussel said of our campaign:

It is outrageous that a mandatory submission platform can change their terms and propose using our hard work to train their AI models… This is not what we signed up to, nor do we consent to this.

She’s right. You go to university – and pay a hefty fee – to train yourself for a future career. Not AI. If these changes go ahead, you’d have no choice.

A question of ownership

This is urgent for universities to review. If students can’t meaningfully consent to their work being used to train AI, that presents a fundamental challenge to the idea that students own the copyright in their work, whether that’s under the Copyright, Designs and Patents Act or university policy.

Turnitin’s response has been that students would “retain ownership of their submissions” and that work would be used in anonymised or de-identified form. But anonymisation deals with data protection. It does nothing about copyright, because copying an essay into a training set is still copying.

Its latest statement goes further, saying no student or staff data has been used to develop “any generative artificial intelligence (AI) models or large language models (LLM),” and that it has “no plans to use student essays in this way.” But it’s worth reading carefully. The denial covers generative AI, while the proposed terms talked about improving or developing AI tools in general. It slides from “student or staff data” to “student essays” between one sentence and the next. And “no plans” describes current intentions rather than offering any contractual guarantee.

And Turnitin’s current agreement already takes a “non-exclusive, royalty-free, perpetual, worldwide, irrevocable license” to students’ work, which covers “improving the quality of the Services generally.”

The University of York’s Regulation 12.2.4.1 limits the licence each student grants to “the purposes of examination of that Student,” and says that includes sharing work with “providers of anti-plagiarism software” – but it doesn’t name the provider. So what happens when Turnitin changes its own terms?

Walking away is possible

Southampton dropped Turnitin citing a responsibility to protect students’ intellectual property, and has shown that Blackboard Assignment with its SafeAssign plagiarism tool is a practical alternative – courses starting from 14 September aren’t using Turnitin at all.

Aberystwyth, which lobbied against the change alongside UCISA, is staying with Turnitin until its contract ends in July 2027, but has told staff who’d prefer not to use Turnitin that they can switch to Blackboard Assignment. Sussex has dropped Turnitin but hasn’t given a reason, and UEA has similarly indicated that it’s moving away.

UC Berkeley, meanwhile, says the AI provisions do not apply to its users because its own contracts bar the use of institutional data with AI – which shows what a university can secure when it writes the protection in first.

And other universities whose SUs I’ve spoken to appear to be giving an ultimatum – reverse these changes or we’re gone. That option of walking away needs to stay on the table.

A bigger conversation

But this debate isn’t really about Turnitin specifically. It’s a wider conversation that we seem unable, or unwilling, to have in the UK. With massive American AI corporations battling each other for competitive advantage – resulting in the overtly dystopian practice of book burning where niche, rare data is being grabbed and destroyed as they run out of data to train on – we need to reckon with real questions over who has a say over how AI is developed, and who doesn’t.

The absurd pace and geopolitical tension of AI development has meant that any traditional idea of consent and fair pay for work – generating academic work is work – has been thrown out of the window in favour of a mad rush for any and all data, consent be damned.

Humans create the data on which AI is trained – OK, technically some of it is synthetic, but that’s a different story – yet they don’t see a lick of benefit from it. Private companies do.

According to the Ada Lovelace Institute, 83 per cent of the public are concerned about public sector bodies (like Universities) sharing their data with private companies to train AI systems, and 50 per cent said they did not feel represented in AI policymaking.

It’s no surprise then that as students continue to worry about AI-driven job losses that a move by a big American tech firm to grab data to train AI is met with extreme opposition.

There’s a perception there that AI is being trained on the very data meant to train graduates for the jobs market – that is being shrunk by the expansion of that very AI. This begs the question – does UK policy protect enough against that?

Academics were the first to be targeted by AI firms seeking rich, niche training data, and they were sold out by their publishers – Taylor & Francis licensed authors’ work to Microsoft for AI training without telling them. Students appear to be the next target.

In Spain, the author’s express permission is required to reproduce assessed work. Romania voids clauses that sign away students’ IP. In England, we only partially meet the intellectual property provision in the European Students’ Union’s Student Rights Charter.

What needs to happen

Universities need to act now, reviewing their Turnitin account settings, the terms they’ve accepted, and their contract end date against their own IP regulations, alongside their SU. They should cost an alternative and assess viability, as Southampton has done.

Regulation needs guarantees modelled on Spain – commercial firms shouldn’t be allowed to reproduce assessed work beyond assessment without meaningful, explained, opt-in consent.

Sector bodies should speak openly about these challenges and where they stand. UCISA has been the channel for universities’ concerns and is now Turnitin’s partner in the redraft, but none of Jisc, Universities UK, or the Office for Students appears to have said anything in public. And the statement gives students no role at all in rewriting terms that will govern their own work.

And we need to have a proper debate about students’ IP and IP more generally in the UK.

We have until 1 September 2027 to think about what student intellectual property actually means, and the redraft is happening now, so universities should be using their seat at the table to get consent and ownership written into it rather than “clarified” afterwards. More broadly, it’s crucial that we grapple with the implications of AI being trained in human generated work without consent or pay.