Navigating cognitive dissonance in generative AI adoption: Ethical and psychological dimensions of knowledge worker acceptance

Purpose This article examines how knowledge workers manage generative AI adoption in ethically charged workplace contexts, focusing on how perceived usefulness and necessity coexist with ethical and societal concerns. Design/methodology/approach We use a multi-phase mixed-method design combining qualitative interviews, a quantitative survey, and supplementary qualitative follow-up interviews. Study 1 draws on 22 semi-structured interviews and abductive, theory-informed qualitative analysis, informed by constructivist grounded theory principles, to develop a process account of acceptance under moral tension. Study 2 surveys 543 knowledge workers and uses latent-variable structural equation modelling to examine selected relationships suggested by the qualitative mechanism. Findings Study 1 shows that acceptance is shaped by tension between “must-use” necessity beliefs and ethical/societal risk beliefs. Knowledge workers manage this tension through strategies such as input sanitization, selective “safe-task” use, verification routines, responsibility shifting, trivialization, and bolstering. Study 2 provides partial quantitative support: perceived risk and perceived necessity strongly predict experienced cognitive dissonance, while the dissonance–intention relationship is specification-sensitive, significant in the direct model but non-significant in the mediated specification. The high HTMT value between perceived usefulness and behavioural intention further requires caution in interpreting PU-related paths. Originality/value The study contributes a mechanism-oriented account of generative AI adoption under ethical tension. Cognitive dissonance is theorized as a complementary process lens, not as a validated replacement for TAM3.