This study examines the effects of artificial intelligence (AI) mediation on students’ psychological safety, empathic engagement, and understanding of historical and literary texts in the context of teaching Zar Zaman poetry. The theoretical framework integrates the multidimensional model of empathy, the concepts of psychological safety and self-efficacy, and the sociocultural approach to development. The study involved 60 eighth-grade students from two schools in Almaty, Kazakhstan. Using a quasi-experimental design, the participants were assigned to a control group (n=30) and an experimental group (n=30). During the four-week intervention, the control group studied Zar Zaman poetry using traditional instructional methods, whereas the experimental group received traditional instruction supplemented by individual dialogues with generative AI and culturally and historically contextualized prompt engineering. Psychological changes were assessed using the Perspective-Taking and Fantasy subscales of an adapted version of Davis’s Interpersonal Reactivity Index (IRI), a scale measuring anxiety during interaction with AI, and semi-structured interviews. The results indicated a 23% increase in perspective-taking and a 21% increase in fantasy identification in the experimental group, compared with increases of 6% and 5%, respectively, in the control group. Anxiety associated with AI interaction decreased by 42% in the experimental group. Rubric-based assessment of text-analysis skills showed a 21% improvement in the experimental group compared with 6% in the control group. Qualitative analysis revealed tendencies toward greater freedom in expressing personal opinions and emotional reactions during AI-mediated dialogue, increased engagement with the perspective of historical subjects, and stronger connections between the literary text and students’ personal experiences. The findings suggest that AI mediation may provide an additional reflective space that can reduce evaluation-related anxiety and support empathic engagement with historically and culturally situated literary texts. However, given the limited sample size and the characteristics of the psychodiagnostic instruments employed, causal interpretations should be made with caution, and the identified relationships require further examination in larger samples.