Does Trust Spill Over?: Exploring Asymmetric Trust Dynamics Between Teammate AI and Guide AI in Cooperative Games

Cooperative games increasingly feature multiple AI agents serving distinct roles, yet how players form and transfer trust across such agents remains unclear. We investigated trust spillover between a task-executing Teammate AI and an advisory Guide AI in a web-based cooperative cooking game (within-subjects, N = 23). Teammate AI ability (high vs. low) and Guide AI accuracy (high vs. low/absent) were experimentally manipulated. Results revealed a directional asymmetry in trust spillover. Teammate AI ability significantly influenced trust in the Guide AI ( $\eta _p^2 =.220$ ), whereas Guide AI accuracy and presence had no effect on trust in the Teammate AI ( $\eta _p^2 =.002$ ). This suggests that, under cognitive constraints, trust propagates from high-salience task agents to lower-salience advisory agents, but not vice versa. Additionally, the presence of the Guide AI improved performance and perceived control independent of advice accuracy, suggesting an attentional cueing effect. Finally, congruent agent quality levels enhanced perceived relatedness, highlighting the importance of system-level coherence. These findings demonstrate asymmetric trust spillover in cooperative games with role-differentiated AI, where execution-focused agents anchor system-wide trust formation, with implications for multi-agent AI systems in which agents serve distinct functional roles.