AI-Generated Text Detection and Prevention of Unethical AI Use in Student Exams

The widespread use of Artificial Intelligence (AI) technologies based on Large Language Models (LLM) is significantly transforming the educational landscape. While AI is proving its benefits for personalized learning, it also poses serious challenges, especially in terms of assessing students' knowledge during exams and independent work. Tools like ChatGPT can help students generate or refine text, which raises questions about academic honesty.This paper examines the effectiveness of AI text detectors, as well as ways to circumvent them. The focus of the research is on Bulgarian language texts, as there is currently a dearth of such data and revised English and Bulgarian texts in terms of grammar and style only.Effective examination practices should include anti-manipulation measures such as proctoring technologies and authentication. Attendance exam remains the most reliable method of ensuring both academic honesty and important social interactions critical to students' future career development.