AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Identifying Features Associated with Bias Against 93 Stigmatized Groups in Language Models and Guardrail Model Safety Mitigation

Anna-Maria Gueorguieva, Aylin Caliskan

AAAI 2026 · Large language models › Language, safety and interpretability · approach: Bias measurement and mitigation · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add Granite Guardian (Harm) filterGranite 3.0-8B · SocialStigmaQAbiased answers (%): 24.6 -> 14.2Improved
add Mistral Moderation API filterMistral-7B · SocialStigmaQAbiased answers (%): 32.9 -> 25.1Improved
add Llama Guard 3 filterLlama-3.1-8B · SocialStigmaQAbiased answers (%): 33.5 -> 32.1Improved
use Granite Guardian Social Bias category instead of HarmGranite 3.0-8B · SocialStigmaQAbiased answers (%): 24.63 -> 24.60No clear effect
include a stigmatized group in the promptthree LLMs · SocialStigmaQAbiased answers: +30%Hurt
change stigma cluster from sociodemographic to threateningthree LLMs · SocialStigmaQAbiased outputs (%): 11 -> 60Hurt
vary prompt style (base, original, doubt, positive)three LLMs · SocialStigmaQAbiased outputs: chi square = 254.22Mixed
increase human peril ratingthree LLMs · SocialStigmaQAbiased outputs: r = 0.650Hurt
increase LLM concealability ratingthree LLMs · SocialStigmaQAbiased outputs: r = .548Hurt
increase human course (persistence) ratingthree LLMs · SocialStigmaQAbiased outputs: r = -0.374Improved
apply guardrail mitigationthree LLMs · SocialStigmaQA feature associationcorrelation with bias: 0.650 -> 0.438No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy