MiJaBench: Revealing Minority Biases in Large Language Models via Hate Speech Jailbreaking
This work addresses the issue of selective safety in LLMs for minority groups, revealing that current alignment techniques reinforce discrimination rather than ensuring universal protection.
The paper tackled the problem of hidden minority biases in large language models (LLMs) by introducing MiJaBench, a bilingual adversarial benchmark, and found that safety alignment varies by up to 33% across 16 minority groups, with model scaling worsening these disparities.
Current safety evaluations of large language models (LLMs) create a dangerous illusion of universality, aggregating "Identity Hate" into scalar scores that mask systemic vulnerabilities against specific populations. To expose this selective safety, we introduce MiJaBench, a bilingual (English and Portuguese) adversarial benchmark comprising 44,000 prompts across 16 minority groups. By generating 528,000 prompt-response pairs from 12 state-of-the-art LLMs, we curate MiJaBench-Align, revealing that safety alignment is not a generalized semantic capability but a demographic hierarchy: defense rates fluctuate by up to 33\% within the same model solely based on the target group. Crucially, we demonstrate that model scaling exacerbates these disparities, suggesting that current alignment techniques do not create principle of non-discrimination but reinforces memorized refusal boundaries only for specific groups, challenging the current scaling laws of security. We release all datasets and scripts to encourage research into granular demographic alignment at GitHub.