Proceedings of the International scientific and practical conference “Science, Technology and Culture: Strategies for Sustainable Development” (December 15-17, 2025) / Publisher website: www.naukainfo.com. – Krakow, Poland, 2025. – 120 p.

75 prosody and accent), grammatical accuracy, and fluency. Some tools also offer explicit explanations as to why particular expressions may sound awkward or unnatural. Furthermore, the system is capable of designing an individualized learning pathway by identifying users’ needs and goals at a remarkably deep level. Frequent speaking practice is widely regarded as a key prerequisite for achieving oral fluency. In the domain of writing, LLM-based assistants (e.g., Paperpal or GrammarlyGO ) have moved far beyond basic spell-checking, employing natural language processing (NLP) and machine learning to conduct comprehensive text analysis and to identify issues related to structure, tone, and clarity. Their generative capabilities provide advanced support that is particularly valuable for non-native English-speaking (NNES) researchers. LLMs can deliver immediate formative feedback by suggesting strategies to enhance coherence, lexical diversity, and adherence to a formal academic register. Generative functions further enable improved word choice, increased fluency through paraphrasing, and the provision of accurate academic translation. Despite its advantages, the integration of generative AI entails significant risks that must be addressed within educational contexts. First, there is a risk of overreliance on technology. Scholars caution that unguided use of LLMs to obtain ready-made answers may undermine long-term knowledge acquisition. In addition, LLMs are prone to generating inaccurate information, commonly referred to as “hallucinations.” As these models are unable to independently verify the factual accuracy of their outputs, effective educational use of AI requires mandatory critical evaluation of generated content by learners. Second, one of the most acute socio-ethical challenges is systemic bias against non-native English speakers (NNES). Empirical studies have shown that more than half of academic writing samples produced by NNES were falsely classified as AI- generated by GPT detection tools, whereas detection accuracy for native speakers was nearly perfect. This issue arises because NNES writing, in striving for formal correctness, often exhibits lower textual perplexity (i.e., greater predictability for language models), which is erroneously interpreted as machine-generated. Such