Beyond Performance: Rethinking Trust, Responsibility, and Hallucination in AI-Assisted Decision-Making

Recent advances in artificial intelligence, particularly large language models, have accelerated the adoption of AI-assisted decision-making in organizational and societal contexts. However, dominant discussions of trust in AI remain largely performance-oriented, focusing on accuracy and efficiency while overlooking critical socio-technical risks. This paper argues that trust in AI cannot be adequately understood without addressing hallucination, opacity, and emerging responsibility gaps between human and artificial agents. By critically examining existing AI trust frameworks, the study highlights their limitations in accounting for epistemic uncertainty introduced by generative systems. Drawing on interdisciplinary literature, this paper develops a conceptual framework that reconceptualizes trust as a conditional and relational construct shaped by human judgment and accountability mechanisms.