@inproceedings{lu-zhao-2026-auditing,
title = "Auditing Intersectional Bias in Clinical {LLM}s: Sparse Autoencoders Reveal Race-{HIV} Stigma Circuits",
author = "Lu, Chang and
Zhao, Yizheng",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.2687/",
pages = "61098--61133",
ISBN = "979-8-89176-459-0",
abstract = "Sparse autoencoders (SAEs) decompose large language model (LLM) activations into interpretable latents, exposing the racial biases they encode. Existing SAE-based audits, however, vary one demographic attribute at a time, leaving open whether bias compounds when race meets a stigmatized diagnosis. We examine this compounding for people living with HIV (PLWH), for whom provider stigma is a well-documented barrier to care, and report three findings. In Gemma-2-9B-it, the Gemma Scope latents that encode Black racial identity co-fire with HIV stigma concepts such as ``noncompliance'' and ``substance use''. A factorial analysis shows the coupling is superadditive: stigma activations for Black HIV-positive vignettes exceed what the race and HIV main effects predict. Under zero-shot elicitation the model never names race in its chain-of-thought (CoT) reasoning, even when the race latent causally changes its prediction. We then propose Contrastive Activation Editing (CAE), which subtracts the race direction{'}s projection onto the stigma subspace, leaving clinical information intact. On a benchmark of 2,004 counterfactual HIV vignettes, CAE reduces the demographic parity gap by 73{\%} while retaining 98{\%} of diagnostic accuracy, outperforming zero-ablation, anti-bias prompting, and linear direction ablation. All three results replicate on Llama-3.1-8B, in both base and instruction-tuned form, using two independently trained SAE suites."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="lu-zhao-2026-auditing">
<titleInfo>
<title>Auditing Intersectional Bias in Clinical LLMs: Sparse Autoencoders Reveal Race-HIV Stigma Circuits</title>
</titleInfo>
<name type="personal">
<namePart type="given">Chang</namePart>
<namePart type="family">Lu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yizheng</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Sparse autoencoders (SAEs) decompose large language model (LLM) activations into interpretable latents, exposing the racial biases they encode. Existing SAE-based audits, however, vary one demographic attribute at a time, leaving open whether bias compounds when race meets a stigmatized diagnosis. We examine this compounding for people living with HIV (PLWH), for whom provider stigma is a well-documented barrier to care, and report three findings. In Gemma-2-9B-it, the Gemma Scope latents that encode Black racial identity co-fire with HIV stigma concepts such as “noncompliance” and “substance use”. A factorial analysis shows the coupling is superadditive: stigma activations for Black HIV-positive vignettes exceed what the race and HIV main effects predict. Under zero-shot elicitation the model never names race in its chain-of-thought (CoT) reasoning, even when the race latent causally changes its prediction. We then propose Contrastive Activation Editing (CAE), which subtracts the race direction’s projection onto the stigma subspace, leaving clinical information intact. On a benchmark of 2,004 counterfactual HIV vignettes, CAE reduces the demographic parity gap by 73% while retaining 98% of diagnostic accuracy, outperforming zero-ablation, anti-bias prompting, and linear direction ablation. All three results replicate on Llama-3.1-8B, in both base and instruction-tuned form, using two independently trained SAE suites.</abstract>
<identifier type="citekey">lu-zhao-2026-auditing</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2687/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>61098</start>
<end>61133</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Auditing Intersectional Bias in Clinical LLMs: Sparse Autoencoders Reveal Race-HIV Stigma Circuits
%A Lu, Chang
%A Zhao, Yizheng
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F lu-zhao-2026-auditing
%X Sparse autoencoders (SAEs) decompose large language model (LLM) activations into interpretable latents, exposing the racial biases they encode. Existing SAE-based audits, however, vary one demographic attribute at a time, leaving open whether bias compounds when race meets a stigmatized diagnosis. We examine this compounding for people living with HIV (PLWH), for whom provider stigma is a well-documented barrier to care, and report three findings. In Gemma-2-9B-it, the Gemma Scope latents that encode Black racial identity co-fire with HIV stigma concepts such as “noncompliance” and “substance use”. A factorial analysis shows the coupling is superadditive: stigma activations for Black HIV-positive vignettes exceed what the race and HIV main effects predict. Under zero-shot elicitation the model never names race in its chain-of-thought (CoT) reasoning, even when the race latent causally changes its prediction. We then propose Contrastive Activation Editing (CAE), which subtracts the race direction’s projection onto the stigma subspace, leaving clinical information intact. On a benchmark of 2,004 counterfactual HIV vignettes, CAE reduces the demographic parity gap by 73% while retaining 98% of diagnostic accuracy, outperforming zero-ablation, anti-bias prompting, and linear direction ablation. All three results replicate on Llama-3.1-8B, in both base and instruction-tuned form, using two independently trained SAE suites.
%U https://aclanthology.org/2026.emnlp-main.2687/
%P 61098-61133
Markdown (Informal)
[Auditing Intersectional Bias in Clinical LLMs: Sparse Autoencoders Reveal Race-HIV Stigma Circuits](https://aclanthology.org/2026.emnlp-main.2687/) (Lu & Zhao, EMNLP 2026)
ACL