If it doesn’t have a soul, it doesn’t hurt? The effect of bullying and social anxiety on the use of conversational AI

Introduction: Conversational artificial intelligence (CAI) increasingly functions as a social interaction partner, raising questions about how established interpersonal behaviors transfer to human-AI interaction. The present study examined whether bullying-like behavior directed toward CAI reflects the extension of real world social strategies, and how social anxiety, prior bullying experiences, and AI self-competency contribute to such behavior. Methods: = 289) completed measures of social anxiety, prior bullying perpetration and victimization, AI self-competency, and bullying-like behavior directed toward conversational AI. Results: Through several analyses, bully-victims exhibited the highest levels of AI-directed bullying-like behavior. The interaction between bullying perpetration and victimization emerged as the strongest predictor in our regression analysis, while prior bullying perpetration alone was not sufficient to explain AI-directed aggression, and prior victimization alone was negatively associated with AI-directed bullying-like behavior. Social anxiety did not directly predict bullying-like behavior toward AI; however, mediation analyses showed that prior victimization fully mediated the relationship between social anxiety and bullying-like behavior towards CAI. Individuals with higher social anxiety were more likely to report victimization, which in turn was associated with increased bullying-like behavior toward AI. AI self-competency did not emerge as an independent predictor. Discussion: These findings suggest that AI-directed bullying-like behavior is best understood as a context-dependent behavior arising from the co-occurrence of aggression and vulnerability, rather than a direct extension of general aggressive tendencies. Conversational AI constitutes a distinct relational context that can both reflect and reinforce existing behavioral patterns. This has important implications for understanding how human-AI interactions may shape social behavior and psychological functioning over time.