In July 2026, Cambridge University became the first in the world to refuse to sign a new license agreement with the plagiarism detection system Turnitin: the amended Terms of Use of the new EULA allowed the use of student work to train AI models. Under pressure from its most prominent client, the company delayed the release of the new EULA to September 2027, and the University of Southampton announced a complete withdrawal from Turnitin after the 2026–27 academic year. The dispute showed that rights to educational data in EdTech are ceasing to be a silent formality of a user agreement and are becoming a subject of direct negotiations between universities and vendors.

image

What happened

Cambridge continues to use Turnitin under the existing old terms until July 2027 and is simultaneously looking for alternatives to the system. The official Turnitin Trust Center records a key distinction: the company states that it does not train generative AI models or LLMs, but acknowledges fine-tuning of the AI-text detection classifier supplied in the Turnitin Originality license on anonymized student data. According to the student newspaper Varsity, in addition to Southampton, York, St Andrew's, Lancaster, Reading, and East Anglia are considering changing the terms with the vendor.

Context

Turnitin remains the de facto standard for plagiarism checking in British higher education: according to Uni Compare, the company's products are used by 98% of the country's universities. Such dominance meant that terms of use were signed as a silent formality, and the question of training models on student texts was not brought into negotiations and was not discussed publicly. The acuteness of the situation is also determined by the composition of the data: student work is, in fact, an annotated corpus of "human text vs. AI text" pairs, a key resource for training detectors of generated content. The new position of the universities for the first time questions both this resource and the habit of accepting vendor terms without discussion.

Why this matters for the industry

For the EdTech industry, this is a precedent of client negotiating power: Turnitin's concession was won through university pressure, not technical argumentation, as the parties never presented an open independent assessment of the detector's quality. For competitors and startups, the signal is early but specific: the incumbent's procurement lock-in has been shaken, "clean data" and public Trust Centers are turning from a feature into a positioning category, and discussions in universities may grow into RFI/RFPs for replacement, where requirements for training on student data and requests for public detection metrics will appear directly in procurement documents. Another consequence is expected: the "classifier" formulation from the Trust Center will begin to migrate to the licenses of other EdTech companies, as it allows declaring good faith without giving up training on user data.

Why this matters for users

For researchers in AI-text detection, the event provides a rare benchmark: there is a publicly acknowledged industrial classifier trained on a real corpus of student work, but without open metrics — neither the error rate, nor robustness to paraphrasing, nor behavior in different languages and genres have been published. For students and teachers, the dispute is a reminder that the fate of their work is determined by the text of the contract signed by the university: as long as the clause on training on student data is not brought into negotiations, the use of texts is possible by default. A practical step for technical teams and administrators right now is to check what is written in the licenses of your vendors about training models on your data, and to include this clause in contract negotiations.

What is still unknown / limitations

The key unknown is the quality of detection: there is no public benchmark or description of the classifier's validation in the sources, so it is currently impossible to verify the vendor's claims about accuracy. The specific alternatives to Turnitin that Cambridge is considering and their terms are not named in the sources. There are currently two confirmed decisions — Cambridge's position and Southampton's exit, while the involvement of York, St Andrew's, Lancaster, Reading, and East Anglia remains at the discussion level. The material is based on a Varsity article and the official Turnitin Trust Center pages, and discussion of the news on Hacker News is virtually absent.

Sources

Author

Look at AI, editorial team