Rethinking Ethical Responsibility and Data Governance in Academic Assessment Using Large Language Models

Background: The integration of Large Language Models (LLMs) into academic grading practices has expanded rapidly in higher education, driven by demands for efficiency and consistency. Aims: In response to these concerns, this study seeks to explore issues of ethical accountability and data governance in the use of LLMs for academic assessment, drawing on the perspectives of lecturers, students, and academic administrators. Methods: The study adopted a qualitative exploratory approach to capture in-depth insights into current assessment practices involving LLMs. Data were gathered through semi-structured interviews, institutional document analysis, and direct observations across selected higher education institutions. Analysis followed the interactive framework proposed by Miles, Huberman, and Saldaña, involving iterative processes of data reduction, data display, and conclusion verification, with triangulation applied to strengthen trustworthiness. Result: The findings demonstrate a set of interrelated challenges. The involvement of LLMs in grading processes often obscures responsibility for assessment decisions, particularly when transparency is limited. Concerns regarding fairness and potential bias persist, especially in evaluating varied linguistic and contextual expressions. At the same time, data governance mechanisms remain insufficiently developed, with unclear procedures for consent, data storage, and regulatory compliance. These issues collectively reflect uneven institutional preparedness and weak ethical oversight. Conclusion: The study concludes that the use of LLMs in academic grading requires clearly defined ethical accountability and comprehensive data governance frameworks. Continued human oversight, supported by institutional policies and capacity-building initiatives, is essential to safeguard academic integrity and ensure responsible adoption of AI-assisted assessment in higher education.