2026
doi: 10.12720/jait.17.6.1084-1095
Countering Paraphrasing and Style Transfer in Educational Settings: A GAN-Based Approach to Identifying AI-Generated Academic Essays
- IME Team, MASI Laboratory, Mohammadia School of Engineers (EMI), Mohammed V University in Rabat, Rabat, Morocco
- Manuscript receivedDecember 23, 2025
- revisedJanuary 14, 2026
- acceptedMarch 31, 2026
- publishedJune 10, 2026
Abstract
The generation of content by Artificial Intelligence (AI) in educational settings represents a growing consequence of the irresponsible use of generative AI that threatens academic integrity. Mitigating this phenomenon through AI-based detection remains a major challenge, given that generative AI tools, such as Large Language Models (LLMs), evolve rapidly. Classical detection tools such as anti-plagiarism software are no longer capable of tracking or curbing the spread of this phenomenon. Furthermore, the existing AI-based tools fail to detect the content generated, paraphrased, or humanized by bots that alter sentence structure. In this article, we propose a novel architecture based on Generative Adversarial Networks (GANs). This framework is a generative model composed of two main components trained through a minimax game. The first is the generator; it generates or reformulates text, while its opponent, the discriminator, tries to detect whether the text is human-generated or produced by the generator. We used DistilGPT-2 as the generator, trained simultaneously with DistilRoBERTa, the discriminator. Our framework was evaluated on 44,868 academic text samples and achieved 99% precision on unmodified texts. Moreover, our approach demonstrates strong reliability, maintaining a precision of 92% despite paraphrasing and humanization, as well as noise injection attacks that render other detection tools ineffective. The use of GANs enables our discriminator to more effectively identify deep semantic patterns and reformulated structures. Finally, we discuss future research directions to improve the adaptability and long-term relevance of the proposed approach in light of rapid advances in artificial intelligence.
Keywords
- education
- Generative Artificial Intelligence (GenAI)
- Generative Adversarial Network (GAN)
- paraphrasing attacks
- academic integrity
- adversarial training
- reinforcement learning
- AI learning
How to Cite
Zohair Elmourabit and Asmaâ Retbi, "Countering Paraphrasing and Style Transfer in Educational Settings: A GAN-Based Approach to Identifying AI-Generated Academic Essays," Journal of Advances in Information Technology, vol. 17, no. 6, pp. 1084-1095, 2026. https://doi.org/10.12720/jait.17.6.1084-1095
Copyright & License
Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).






