Sharing blame with me: How AI anthropomorphism encourages individual unethical behavior

Anthropomorphism is increasingly incorporated into AI agent design to enhance human–AI interaction. However, prior research has largely emphasized its benefits, leaving unclear how anthropomorphic AI shapes moral behavior when AI actively participates in decision-making. Across six studies (N = 1680), including a real-time human–AI interaction experiment, we show that highly (vs. lowly) anthropomorphized AI increases individuals’ tendency to engage in unethical behavior. We identify a responsibility-attribution process underlying this effect: AI anthropomorphism increases perceived AI moral agency, which shifts perceived responsibility for decision outcomes from the self toward AI and, in turn, increases unethical behavior. Further, this effect is attenuated when personal responsibility is made salient or when AI service psychological ownership is high. These findings reveal a potential dark side of AI anthropomorphism by showing how humanlike AI design reshapes moral agency perceptions and responsibility attribution, with implications for the responsible design and use of AI agents.