Negotiating generative AI use and academic integrity in Saudi EMI higher education: A qualitative interview study of student and instructor perspectives

For multilingual students completing degrees through English-medium instruction (EMI), generative artificial intelligence (GenAI) tools such as ChatGPT sit in an unclear space between legitimate second-language (L2) support and the substitution of the linguistic and cognitive work that assessment is designed to develop. This qualitative study draws on semi-structured interviews with five undergraduate EFL students and six instructors teaching in EMI programmes at one Saudi university to examine how both groups conceptualise and negotiate the boundary between GenAI-mediated language support and academic misconduct. Reflexive thematic analysis was framed by three established theories used in combination as complementary interpretive lenses: the ICAI values framework, Neutralization Theory and the Technology Acceptance Model, each operationalised as a distinct deductive coding domain. A “grey zone'' was identified in which the acceptability of GenAI use for L2 academic writing remained deeply uncertain: students valued GenAI as linguistic scaffolding (phrasing, grammar and vocabulary support that echoes earlier debates over machine translation) while simultaneously fearing dependency and the loss of an emerging academic voice in English. Instructors reported low confidence in detecting AI-mediated writing, articulated a risk we term perceived academic hollowing, that is, the feared erosion of the linguistic and disciplinary competencies an assessment is designed to develop, and judged existing integrity policies inadequate for L2 writers. Both groups called for explicit guidelines, critical AI literacy instruction and process-oriented assessment of language and learning. All findings report what participants perceived and reported; no learning outcomes, assignment artefacts or records of AI interaction were measured. The study therefore offers analytic insight into how the language support-substitution boundary is negotiated in one EMI setting, and transferability to comparable contexts rather than statistical generalisation or evidence of educational effect.