Research Summary: ABLEIST: Intersectional Disability Bias in LLM-Generated Hiring Scenarios

A research audit of six large language models across 2,820 hiring scenarios evaluated intersectional disability bias, focusing on understudied non-Western contexts involving caste, gender, and nationality. The study introduces the ABLEIST framework, establishing five ableism-specific and three intersectional harm metrics to quantify biases such as ableism, inspiration, superhumanization, and tokenism. Note that the source abstract is partially truncated at its conclusion.

Why it matters

This research matters because organizations increasingly deploy LLMs in recruitment workflows without evaluating non-Western intersectional biases, creating legal, reputational, and ethical exposures. It highlights that standard AI bias assessments fail to capture specific forms of marginalization affecting people with disabilities across international labor markets.

Key insights

• Large language models (LLMs) perpetuate identity-based discrimination in high-stakes hiring scenarios, particularly affecting people with disabilities (PwD).
• Existing AI fairness research is predominantly Western-centric, overlooking intersectional marginalization factors such as caste and gender in the Global South.
• An empirical audit was conducted across six LLMs covering 2,820 hiring scenarios featuring diverse disability, gender, nationality, and caste profiles.
• The study proposes the ABLEIST framework, defining five ableism-specific and three intersectional harm metrics to capture nuances including ableism, inspiration, superhumanization, and tokenism.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2510.10998