Research on how people judge AI compared to humans remains fragmented despite its criticalityin resolving a multitude of current AI ethical debates. This pre-registered meta-analysis (OSF:10.17605/OSF.IO/GUHPQ; PROSPERO: CRD420251237117) addresses this by systematically reviewingexperimental evidence from 50 experimental studies (363 comparisons) comparing moral judgments ofhuman and AI agents’ decisions in moral dilemmas (e.g. the trolley problem) which were extracted fromPubMed, Web of Science and Elicit. Standardized effect sizes (Hedge’s g) were computed for two-celldesigns (Agent: Human vs. AI). Risk of bias was assessed using a pre-registered checklist for experimentaldesigns. Results from a three-level random-effects meta-analysis show that AI agents’ moral decisions arejudged as less morally permissible than equivalent human decisions (g = -0.08, p = 0.035). Among severalmoderation analyses, perceived mental capacity (PMC) – ascribed through proxies such as agency andexperience – significantly reduced and reversed the negative effect (g = 0.11, p = 0.007). Higher PMCcorresponded to smaller or reversed differences between moral evaluations of human and AI decisions.Interpretability was constrained by substantial heterogeneity across designs. Together, these findingsreveal asymmetries in the moral frameworks guiding evaluations of human and AI agents and point topsychological factors that may contribute to variability in human-AI moral expectations. No externalfunding.