From Myth to Model: Representation of “The Jew” in Generative AI

Antisemitism, a persistent form of cultural prejudice, is increasingly manifesting in subtle and modernized forms within contemporary society. This study examined how cultural representations of Jews are encoded in large language models (LLMs), which absorb societal discourse through training on vast corpora of human-generated text. To investigate these representations, we developed an indirect methodology: generating character biographies using Jewish and non-Jewish names, removing overt identity markers, and prompting the LLM to evaluate a range of psychological and sociocultural traits. Characters associated with Jewish (vs. non-Jewish) names were consistently rated as more competent, privileged, dominant, self-controlled, future-oriented, hierarchical, and obsessive—and as less likable and less collectivistic. When these trait profiles were mapped onto fictional characters, they aligned with archetypes historically associated with antisemitic tropes. These findings demonstrate how latent stereotypes can persist in AI systems despite explicit bias mitigation, highlighting the adaptability of prejudice and the structural embedding of cultural archetypes in contemporary technologies.