A Trust-Centered Conceptual Model for AI-Driven Multilingual Emergency Communication

Trust is the hinge on which emergency warning response turns: a message that is received, translated, and rendered accessible still fails if its recipient doubts its authenticity or its source. Artificial intelligence (AI)-driven multilingual alerting complicates the trust judgment by inserting an automated, cross-linguistic layer between the issuing institution and the recipient, requiring recipients to extend trust not only to a human authority but to an automated process. This article develops a trust-centered conceptual model of AI-driven multilingual emergency communication. Integrating crisis and emergency risk communication, source-credibility theory, the literature on trust in automation, and cross-cultural and linguistic mediation, the model specifies five antecedents (translation fidelity, cultural and linguistic adaptation, AI transparency and disclosure, perceived human oversight, and prior institutional trust), three mediating mechanisms (perceived source credibility, message comprehension, and uncertainty reduction), and two behavioral outcomes (trust in the alert and sender, and protective action and compliance). Five moderators (language proficiency, cultural background, digital access and literacy, disability status, and prior emergency experience) condition the strength of these relationships, formalizing the heterogeneity of warning response and treating equity as a structural property of the model. Twelve propositions are advanced, including the prediction, drawn from the asymmetry of automation trust, that a single salient error damages trust more than an equivalent success builds it, implying that AI-alerting programs are more fragile than their average performance suggests. The article discusses implications for theory, for system design, and for a staged program of empirical testing.