Abstract

Generative Artificial Intelligence has rapidly become an important source of information communication and knowledge production yet its outputs may reproduce social inequalities embedded within training data and cultural representations. In India caste provides a particularly significant context for examining algorithmic bias because social identity remains closely connected with historical inequalities in education occupation representation and social status. This paper comparatively examines caste-related representation in generative AI by synthesising recent research on large language models Indian social identities and caste stereotypes. It focuses on representation bias association bias occupational stereotyping and differential treatment of marginalised and dominant caste groups. The study adopts a comparative socio-technical framework to examine how different dimensions of caste representation appear across AI-generated outputs. The analysis indicates that generative AI can repr