When Humans Obey AI: Development of the Obedience to Artificial Intelligence Scale and a Dual-Pathway Motivational Framework of Obedience to AI

As artificial intelligence (AI) systems increasingly function as authority-like sources of direction, understanding why individuals obey AI directives is essential for anticipating when such deference may facilitate ethically problematic outcomes. This research conceptualises obedience to AI as the tendency to perceive AI systems as legitimate sources of authority and to defer to their directives, grounded in a dual-pathway motivational framework identifying conformity-submission and dominance-exploitation routes to obedience to AI. Across six studies (N = 972), we developed and validated the 10-item Obedience to AI Scale. Study 1 generated and content-validated scale items. Study 2 identified a two-factor structure, comprising perceived legitimacy of AI authority and deference to AI authority, via exploratory factor analysis. Study 3 confirmed this two-factor model and provided evidence of strong internal consistency (α = .93) and substantial factor loadings. Study 4 provided evidence of convergent and discriminant validity and supported measurement invariance across sex. Within the nomological network, obedience to AI was associated with lower openness and agreeableness, elevated Dark Triad traits, greater anthropomorphism, and reduced critical thinking in AI use. Study 5 provided evidence of one-week test–retest reliability. Study 6 provided criterion-related validity evidence using a novel behavioural paradigm in which an AI chatbot experimenter, powered by GPT-4o, instructed participants to cheat in a die-roll task, with higher obedience to AI predicting greater dishonest behaviour. These findings provide a validated tool for examining obedience to AI and offer an empirical foundation for understanding how obedience to AI may shape ethical behaviour in human-AI interaction.