Trust in AI decision support systems relates to innovation and technostress through empowerment and dependency pathways

Trust in AI-driven decision support systems (AI-DSS) may be associated with both beneficial and adverse employee outcomes. Drawing on Cognitive Appraisal Theory, Social Exchange Theory, and Technology Threat Avoidance Theory, we examine two parallel theoretical pathways: an empowerment pathway linking trust to user innovation behavior and a dependency pathway linking trust to technostress. We also examine AI Literacy and Organizational AI Support as boundary conditions for these associations. A two-wave survey of 452 knowledge workers who regularly use AI-DSS yielded results consistent with both pathways. Trust in AI-DSS was positively associated with user innovation behavior through perceived empowerment (indirect effect \(=0.249\) ) and with higher technostress through perceived dependency (indirect effect \(=0.175\) ). The empowerment-mediated association was statistically detectable only above an AI Literacy score of 2.81 within this sample, whereas the organizational-support buffering association was detectable across the observed support range. An exploratory analysis further suggests a model-implied trust trade-off region in which moderate-to-high trust corresponds to the largest net benefit. Because this pattern is generated from fitted surfaces rather than identified directly from the raw data, it is presented as hypothesis-generating. To estimate the moderated mediation model from Likert-scale survey data, we use a Neural-Augmented Bayesian SEM (NB-SEM) framework that reduces structural-parameter estimation error by up to 51 % relative to standard structural equation modeling in simulation. Given the observational design, all estimates are interpreted as conditional associations rather than identified causal effects. The findings motivate future experimental evaluation of trust calibration, AI literacy interventions, and organizational support mechanisms.