Abstarct

Spoken interaction plays a central role in everyday social and institutional life. This paper examines how interactional disfluency becomes vulnerable to misinterpretation as suspicious or risky within fraud-prevention and identity-verification contexts, particularly in systems that incorporate automated decision-making and AI-supported risk assessment. While existing literature has explored the medical and psychological impacts of stuttering, there is limited insight into how communicative difference is interpreted within technology-mediated fraud-prevention practices. Using a critical documentary and conceptual analytic approach, this paper synthesises literature on stuttering, neurodiversity, and social harm to examine how assumptions about fluent, rapid, uninterrupted speech are embedded within institutional procedures and implicit models of credibility. It identifies procedural, structural, and linguistic barriers experienced by people who stutter (PWS), which can generate communicative friction and expose them to heightened scrutiny during verification processes. In these contexts, speech is often treated as a behavioural signal of credibility or risk, despite limited transparency regarding how such signals are interpreted. The session argues that communicative fluency operates as an implicit proxy for trustworthiness within fraud detection systems, creating a critical explainability gap. From an explainable AI perspective, this highlights how behavioural features may be incorporated into decision-making without sufficient interpretability or consideration of communicative diversity, leading to potential misclassification and structural harm.