From tool to teammate? Negotiating accountability and disclosure in academic generative AI use

Abstract

Generative AI is becoming embedded in academic work, yet ethical boundaries and disclosure expectations remain inconsistently defined across contexts and contested. As these systems increasingly resemble collaborative partners rather than neutral tools, accountability questions become more complex, including what counts as acceptable assistance, who is responsible for errors, and what should be disclosed to different audiences. Drawing on focus groups with 15 university academics, this paper examines how academics in this context negotiate ethical boundaries for generative AI across academic tasks, how moral accountability and professional identity are performed in peer dialog, what transparency and disclosure norms emerge across key audiences, and how perceived competence trajectories may be associated with shifts in both boundary-setting and emotional positioning over time. Findings show that boundary-setting is strongly situational, with acceptability calibrated by task type, stakes, and visibility, and anchored in concerns about authorship, authenticity, responsibility, and credibility. Participants enacted moral accountability through tool reframing, verification, reliance minimization and self-presentational disclosure choices. Transparency norms were conditional and audience-dependent, shaped by uncertainty about institutional guidance and by anticipated judgement from colleagues, students and publishers. Adoption trajectories appeared to be associated with these negotiations: greater familiarity was often associated with increased confidence and more nuanced ethical rules, while lower perceived competence was associated with cautious use and heightened concern. The paper concludes with implications for responsible guidance in higher education, arguing for policy and professional learning that recognize the social dynamics of accountability and support calibrated, context-sensitive transparency.