AI anxiety and AI dependence among undergraduates: a moderated mediation model of AI self-efficacy and AI literacy

Abstract

Background The rapid integration of Artificial Intelligence (AI) in higher education offers transformative learning potential but has concurrently triggered psychological challenges, specifically AI anxiety and AI dependence. While these phenomena are increasingly prevalent, the psychological mechanisms converting affective strain into behavioral reliance remain underexplored. Grounded in Cognitive Behavioral Theory and Conservation of Resources Theory, this study investigates the relationship between AI anxiety and AI dependence among undergraduates. Specifically, it examines whether AI self-efficacy mediates this relationship and whether AI literacy acts as a boundary condition moderating these pathways. Methods A cross-sectional survey was conducted with 400 undergraduates recruited via stratified sampling from a normal university in Southwest China. Participants completed validated scales measuring AI anxiety, AI self-efficacy, AI literacy, and AI dependence. Confirmatory factor analysis was performed to verify instrument validity. Hypotheses were tested using the PROCESS macro (Models 4 and 8) to analyze mediation and moderated mediation effects, utilizing bootstrapping techniques with 5,000 resamples to determine the statistical significance of direct and indirect effects. Results The analysis revealed a significant positive association between AI anxiety and AI dependence. AI self-efficacy was found to partially mediate this relationship, indicating that anxiety exacerbates dependence by eroding students’ confidence in their capabilities. Furthermore, AI literacy served as a significant moderator. Results indicated that the negative impact of AI anxiety on AI self-efficacy was stronger for students with higher AI literacy, suggesting that high literacy may sensitize students to competency gaps under emotional strain. AI literacy did not moderate the direct relationship between anxiety and dependence. Conclusion This study identifies AI dependence as a maladaptive coping response to AI anxiety, driven by diminished AI self-efficacy. The findings challenge the assumption that literacy acts solely as a protective buffer, revealing that without emotional regulation, high literacy may intensify the erosion of self- confidence during anxious states. To mitigate maladaptive dependence, higher education institutions should move beyond technical training to adopt holistic strategies. These should include psychological support to build resilience, mastery experiences to foster self-efficacy, and comprehensive curricula that address the emotional and cognitive dimensions of human-AI interaction.