Psychological safety and perceived risk are associated with emergency nurses' intention to use AI-augmented triage systems.

Emergency department overcrowding places sustained pressure on triage workflows and patient prioritization. Artificial intelligence (AI)-augmented triage systems have been introduced to support emergency decision-making, but frontline adoption may depend on both technology-related perceptions and human-organizational conditions. This study examined factors associated with emergency nurses' attitudes and intention to use AI-augmented triage systems, with particular attention to psychological safety and perceived risk. A multi-hospital cross-sectional survey was conducted among 162 frontline triage nurses across nine pilot hospitals in Shanghai between June and August 2025. All participants had at least six months of emergency triage experience and at least three months of actual experience using the AI-augmented triage system. Partial least squares structural equation modelling was used to assess the measurement and structural models. The model explained 57.2% of the variance in attitude and 41.0% of the variance in intention to use. Task-technology fit (β = 0.483, 95% CI [0.387, 0.574]), perceived explainability (β = 0.385, 95% CI [0.280, 0.484]), and psychological safety (β = 0.401, 95% CI [0.294, 0.512]) were positively associated with attitude. Attitude was positively associated with intention to use (β = 0.629, 95% CI [0.526, 0.710]). Perceived risk showed a small negative moderating effect on the association between attitude and intention to use (β = - 0.139, p = 0.039, f2 = 0.029), although the 95% confidence interval included zero. Common method bias was assessed using procedural and statistical checks; however, same-source bias could not be fully ruled out. Emergency nurses' intention to use AI-augmented triage systems was associated with both technology-related perceptions and human-organizational conditions. Psychological safety was positively associated with attitudes toward AI use, while perceived risk may modestly weaken the translation of favourable attitudes into intention to use. Given the cross-sectional and self-reported design, the findings should be interpreted as associations rather than causal evidence. Implementation strategies should address task fit, explainability, accountability boundaries, and psychologically safe human-AI collaboration.