Clinical AI Liability Literacy, Disclosure Confidence, and Adoption Readiness Among Romanian Healthcare Professionals: An Exploratory Single-Center Cross-Sectional Study

Background and Objectives: Errors generated by clinical artificial intelligence (AI) can obscure responsibility and complicate disclosure and incident reporting. This exploratory study examined AI-error liability literacy and its behavioral correlates among Romanian healthcare professionals. Materials and Methods: An exploratory, single-center cross-sectional survey was conducted from February to May 2026 among 109 clinical, managerial, legal/compliance, and IT/data professionals. No a priori sample-size calculation was performed; only moderate or larger effects were detectable. Participants completed a 20-item AI Error Liability Literacy Index and five-point behavioral measures. Analyses included Spearman correlations, k-means clustering, logistic regression, and covariate-adjusted mediation, with Firth penalization, role-adjusted, use-stratified, and competing-model sensitivity analyses, and bootstrap validation. Results: Mean liability literacy was 11.6 ± 4.2/20. Familiarity was higher for the General Data Protection Regulation (GDPR) (67.9%) than for the EU AI Act (24.8%). Literacy correlated with disclosure confidence (ρ = 0.49; 95% CI [0.33, 0.62]), an association that was attenuated but persisted after adjustment for professional role (partial ρ = 0.41), which correlated with adoption intention (ρ = 0.38), as did trust (ρ = 0.43); defensive stance correlated negatively (ρ = −0.31). Three exploratory profiles were identified (bootstrap Jaccard 0.71–0.82) that also differed on variables not used in clustering, and an indirect statistical association between literacy and adoption intention through disclosure confidence was observed (indirect effect 0.12; 95% CI [0.04, 0.22]), although concurrent measurement precludes causal or temporal interpretation. Under penalization, associations for literacy, trust, and disclosure confidence persisted; defensive stance and prior training did not. Conclusions: In this exploratory sample, adoption readiness was associated with regulatory knowledge, disclosure confidence, and calibrated trust. These hypothesis-generating findings require confirmation in larger, longitudinal studies before informing practice, and the appropriate goal is calibrated rather than maximal adoption.