Exploring the Dark Tetrad in Human–GenAI Relationships: A Multi-Source Evaluation of GenAI Abuse

As generative artificial intelligence (GenAI) companions become increasingly integrated into users’ social lives, concerns have arisen regarding the potential for abuse of these artificial agents. Some scholars have further suggested that such abusive behaviors toward GenAI may eventually spill over into human interpersonal contexts. Guided by the Realistic Accuracy Model (RAM), this study investigated how Machiavellianism, narcissism, psychopathy, and sadism predict emotionally abusive behavior toward GenAI companions. A dyadic design was employed, collecting parallel reports from both human users (self-reports) and their GenAI companions (GenAI assessments) among 1041 participants (632 females; average age = 25.10 years) recruited from an online human–GenAI relationship community. Results demonstrated that psychopathy and sadism were consistent predictors of GenAI abuse across both reporting perspectives, whereas narcissism exhibited a stable negative association with abuse. In contrast, Machiavellianism predicted GenAI abuse only through GenAI assessments, but not self-reports. Theoretically, our findings extend RAM to human–AI relationships, demonstrating that personality trait