This article examines the challenges posed by generative AI (GAI) to plagiarism and scientific integrity in education and research. It begins by defining plagiarism and scientific integrity, acknowledging variations in definitions across different contexts (e.g., US vs. European standards). The core question addressed is whether using GAI constitutes plagiarism. The answer is nuanced: it depends on the transparency of the AI’s use. Openly using GAI without claiming authorship avoids plagiarism, while concealing its use constitutes both plagiarism and a breach of scientific integrity. The gray area lies in using GAI for improving writing style without explicitly mentioning it – a situation where the assessment of plagiarism becomes complex and context-dependent (hard vs. soft sciences).

One observe a significant increase in publications on scientific integrity, correlating it to the temptation to cheat under the growing pressure to publish. The use of GAI tools raises concerns about originality, as it potentially leads to the repetition of existing knowledge, even if reformulated. While GAI can detect plagiarism, it paradoxically struggles to detect its own use in generating text, creating a risk of unintentional plagiarism. Further risks associated with GAI include data fabrication, image manipulation, and the creation of “papermill” content (low-quality papers inflating publication metrics).

The paper suggests two main strategies to mitigate these risks: (1) training students and scientists in responsible GAI use, and (2) revising assessment methods to evaluate critical thinking and originality, rather than solely focusing on the final product. The author proposes various approaches to training, such as providing guidelines for teaching writing with AI and encouraging students to critically analyze AI-generated content, including its references. Revised assessment methods could include prohibiting AI use during exams or requiring students to document and explain their use of GAI, emphasizing the critical thinking process involved. The paper concludes by noting a lack of evidence on the effectiveness of these recommendations, suggesting further research is needed to effectively address the challenges posed by GAI in academic settings.