The Same Supply Chain: Plagiarism Detection, Foundation Models, and the Circular Extraction of Student Writing
Abstract: Plagiarism detection systems now feed the generative models they are deployed to detect. In the summer of 2026, Turnitin circulated an updated license agreement claiming the right to harvest student and instructor submissions to train corporate artificial intelligence tools. This expansion marks a backwards form of function creep, transforming tools marketed as defenses against machine generation into extraction pipelines for commercial development. The resulting archive offers immense value because decades of historical student writing provide uncontaminated training corpora essential for averting recursive model collapse. This extraction circumvents individual consent through legal fictions like FERPA's school-official exception, while British institutions challenged the terms under purpose limitation rules. Furthermore, detection classifiers exhibit documented perplexity biases against non-native writers, contaminating the training pool with mislabeled human prose. Education policy treats detection and generation as opposing forces, but in commercial reality they share the same industrial supply chain. Keywords: Plagiarism Detection, Turnitin, Function Creep, Generative Artificial Intelligence, Model Collapse, Training Data Extraction, FERPA, School-Official Exception, UK GDPR, Purpose Limitation, Perplexity Bias, Test of English as a Foreign Language, Academic Integrity, Student Privacy, End-User License Agreement, Higher Education, Uncontaminated Corpora, Bert-Jaap Koops, Ed-Tech Supply Chain, Algorithmic Governance
Authors
- Felix Velasco (ORCID: https://orcid.org/0009-0001-6047-0595)
- Josie Jefferson (ORCID: https://orcid.org/0009-0006-3330-8214)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-15
- DOI
- https://doi.org/10.5281/zenodo.24254205
- Primary Topic
- Academic integrity and plagiarism
- Type
- preprint