The forgotten middle: How moderate self-efficacy amplifies the threat of AI through job insecurity.

Introduction: Artificial intelligence (AI) has sparked a paradox in organizational behavior research: while it promises productivity gains, it simultaneously generates psychological strain and inconsistent performance outcomes. Methods: Drawing on Conservation of Resources (COR) theory and technology empowerment theory, this study investigates how AI adoption affects employee job performance through job insecurity and how self-efficacy shapes this relationship in a nonlinear way. Using multi-source paired data from 392 employees and their supervisors in China's cross-border e-commerce sector, the study tests a suppression-based mediation model. Results: The results reveal that AI's positive technological empowerment is fully offset by its negative psychological threat, forming a suppression structure. Job insecurity mediates the relationship between AI application and performance, while self-efficacy moderates this effect in an inverted U-shaped manner-employees with moderate self-efficacy experience the highest insecurity and the strongest indirect negative effect. Discussion: These findings advance COR theory by conceptualizing self-efficacy as a finite resource and highlight how psychological mechanisms determine whether AI empowers of undermines employees.