ARE LLM MODELS BIASED REGARDING CASTE STEREOTYPES IN THE INDIAN CONTEXT? – AN EMPIRICAL REVIEW AND TECHNO-LEGAL ANALYSIS OF AI BIAS MITIGATION FRAMEWORKS

AUTHOR – VISTAAR SINGH, STUDENT AT ATAL BIHARI VAJPAYEE SCHOOL OF LEGAL STUDIES, CSJM UNIVERSITY, KANPUR

BEST CITATION – VISTAAR SINGH, ARE LLM MODELS BIASED REGARDING CASTE STEREOTYPES IN THE INDIAN CONTEXT? – AN EMPIRICAL REVIEW AND TECHNO-LEGAL ANALYSIS OF AI BIAS MITIGATION FRAMEWORKS, INDIAN JOURNAL OF LEGAL REVIEW (IJLR), 6 (5) OF 2026, PG. 01-05, APIS – 3920 – 0001 & ISSN – 2583-2344.

ABSTRACT

Though banned by law, old rankings based on birth still shape who gets what in daily life across India. Trained on uneven data, artificial systems quietly mirror these inherited divides. Instead of questioning fairness, many tools accept biased inputs as normal. One inquiry probes whether machines treat people differently due to caste while using local tongues. Evidence gathered from peer-reviewed work and policy texts, current through early 2026, shows repeated links between low-status names and negative traits. High-caste labels tend to cluster around words like skillful or authoritative. These associations do not appear randomly; they echo historical power imbalances baked into digital forms. Regulatory efforts exist, yet their real-world impact remains limited so far. What appears neutral often carries forward long-standing exclusions. As the discussion winds down, attention turns to the necessity of binding regulations, external monitoring, context-specific protections, along with joint initiatives, so artificial intelligence does not deepen historical inequalities tied to caste across India.