From Policy to Practice: Academic Leaders’ and Lecturers’ Experiences of Implementing Generative AI Governance and AI-Enabled Assessment Reform in Higher Education

Abstract

Generative artificial intelligence has compelled universities to render generalized promises of integrity, innovation, privacy, and fairness into working guidelines of teaching and evaluation. This secondary qualitative synthesis explores the experiences of academic leaders and lecturers with this translation, what assessment reforms are surfacing, and what organisational conditions should work. Peer-reviewed higher education research published between 2016 and July 2026 was systematically searched. To maintain analytical independence, sources were not used to create the Literature Review. The results were based on a limited number of 15 primary empirical studies published between 2024 and 2026, all in the last five years. Three themes were generated using reflexive thematic synthesis. First, governance is established based on negotiated local discretion; however, discretion is inequitable if common procedural protections are fragile. Second, the shift of assessment reform is chiefly in the direction of examining processes, which can be seen, rather than assessing finished work, such as staged work, oral explanation, reflective disclosure, contextual tasks, and critical assessment of AI output. Third, sustainable implementation requires tuned trust, discipline-related professional learning, approved infrastructure, moderation, and appreciation of redesign workload. The synthesis questions both the prohibition-led governance and the unquestioning adoption. This suggests a tiered architecture where minimum safeguards are specified by institutions, programs understand the requirements of the discipline, and modules declare task-level permissions and evidence. The evidence is still limited in terms of sample size, self-reporting, single institutions, and limited longitudinal assessments. Therefore, the validity of learning, equity, workload, transparency, and procedural justice should be used to evaluate governance, not policy publication or technology acceptance alone.