Applying generative artificial intelligence in higher education: risks, technological limitations, and the distribution of responsibility

Abstract

The article examines the risks, technological limitations, and responsibility models associated with the use of generative artificial intelligence in higher education. The relevance of the topic is determined by the rapid integration of generative AI into teaching, learning, academic writing, assessment, educational content creation, and research support, while the pace of its adoption significantly exceeds the development of coherent pedagogical, ethical, legal, and institutional frameworks. Recent international and national documents increasingly emphasise the need for human oversight, transparency, accountability, academic integrity, protection of human rights, data privacy, and a risk-based approach to the use of AI in education. Therefore, the issue should not be reduced to isolated concerns such as plagiarism or technological convenience. Instead, it requires an integrated analytical model combining risk classification, identification of technological limitations, and a clear distribution of responsibility among the main actors in the educational process. The purpose of the article is to provide a theoretical substantiation of the risks and limitations of generative artificial intelligence in higher education and to develop a model of distributed responsibility for its use among participants in the educational process. The study is based on a set of theoretical methods, including the analysis of scholarly literature, international and national policy and regulatory documents, comparative legal analysis, classification, systematisation, generalisation, and modelling. The study identifies and systematises the main groups of risks related to the use of generative AI in higher education: academic, cognitive, informational, ethical, legal, and organisational risks. It is argued that these risks are interconnected and cannot be reduced solely to issues of academic misconduct. The article separately distinguishes risks from the technological limitations of generative AI systems. The latter include the probabilistic nature of text generation, hallucinations, dependence of output quality on prompt quality, opacity of response formation mechanisms, instability and variability of outputs, and limited contextual and value sensitivity. It is demonstrated that ignoring these limitations often leads to an overestimation of generative AI as a reliable source of academic knowledge rather than a supportive instrument requiring human verification and interpretative control. The article proposes an authorial model of distributed responsibility based on three interrelated levels: the student, the teacher, and the higher education institution. At the individual level, the student is responsible for academic integrity, fact-checking, critical verification, and the proper use of AI tools. At the pedagogical level, the teacher is responsible for assignment design, communicating the acceptable boundaries of AI use, and applying assessment methods that verify understanding rather than merely the ability to generate text. At the institutional level, the higher education institution is responsible for internal policies, methodological guidance, procedural clarity, and the protection of the rights of all participants in the educational process. The study concludes that a productive model for regulating generative AI in higher education is neither total prohibition nor unrestricted use, but rather an institutionally regulated, human-centred, and accountable approach grounded in distributed responsibility and human oversight.