<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="loghmani-etal-2026-hidden">
    <titleInfo>
        <title>Hidden Configuration Effects on Fairness in LLM-Based Multi-Agent Decision Systems</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Hamed</namePart>
        <namePart type="family">Loghmani</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Ninareh</namePart>
        <namePart type="family">Mehrabi</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Homa</namePart>
        <namePart type="family">Hosseinmardi</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Amir</namePart>
        <namePart type="family">Ghasemian</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>Large language models (LLMs) deployed as decision agents in regulated settings—loan approval, hiring, clinical triage—must pass pre-deployment fairness audits that report a single fairness number per model under one implicit configuration. We argue this framing misses the dominant source of variation: configuration itself. Sweeping five axes (persona, interview, deliberation, aggregation rule, order) across three frontier LLMs (Claude Sonnet 4.6, GPT-4o-mini, Grok 4.20 Reasoning) on a 1,000-applicant Adult Census Income sample and a 500-applicant ACSIncome (2024 ACS data) sample, we find that within a single model, persona alone swings positive-prediction rate by up to 91 percentage points and collapses decision rules on two of three models; within a Conservative/Balanced/Progressive panel, replacing the panel chair on identical deliberation transcripts moves the disparate-impact ratio from adverse-impact to near-parity; and reversing the speaking order flips up to 34% of applicants on the same panel. We identify a median-voter effect: the Balanced agent matches the majority-vote verdict on 95.8–100% of applicants across all three models—deliberative aggregation reduces, mechanically, to single-agent decision by the median. The audit unit is the configuration, not the model: a system can pass an audit under one configuration and be silently reconfigured into a non-compliant one through a prompt edit—a change that current regulations do not treat as requiring re-audit. The code to reproduce the experiments in this paper is available at https://anonymous.4open.science/r/agent-decision-bias-emnlp-F059/</abstract>
    <identifier type="citekey">loghmani-etal-2026-hidden</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.2531/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>57565</start>
            <end>57589</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Hidden Configuration Effects on Fairness in LLM-Based Multi-Agent Decision Systems
%A Loghmani, Hamed
%A Mehrabi, Ninareh
%A Hosseinmardi, Homa
%A Ghasemian, Amir
%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 loghmani-etal-2026-hidden
%X Large language models (LLMs) deployed as decision agents in regulated settings—loan approval, hiring, clinical triage—must pass pre-deployment fairness audits that report a single fairness number per model under one implicit configuration. We argue this framing misses the dominant source of variation: configuration itself. Sweeping five axes (persona, interview, deliberation, aggregation rule, order) across three frontier LLMs (Claude Sonnet 4.6, GPT-4o-mini, Grok 4.20 Reasoning) on a 1,000-applicant Adult Census Income sample and a 500-applicant ACSIncome (2024 ACS data) sample, we find that within a single model, persona alone swings positive-prediction rate by up to 91 percentage points and collapses decision rules on two of three models; within a Conservative/Balanced/Progressive panel, replacing the panel chair on identical deliberation transcripts moves the disparate-impact ratio from adverse-impact to near-parity; and reversing the speaking order flips up to 34% of applicants on the same panel. We identify a median-voter effect: the Balanced agent matches the majority-vote verdict on 95.8–100% of applicants across all three models—deliberative aggregation reduces, mechanically, to single-agent decision by the median. The audit unit is the configuration, not the model: a system can pass an audit under one configuration and be silently reconfigured into a non-compliant one through a prompt edit—a change that current regulations do not treat as requiring re-audit. The code to reproduce the experiments in this paper is available at https://anonymous.4open.science/r/agent-decision-bias-emnlp-F059/
%U https://aclanthology.org/2026.emnlp-main.2531/
%P 57565-57589

Markdown (Informal)

[Hidden Configuration Effects on Fairness in LLM-Based Multi-Agent Decision Systems](https://aclanthology.org/2026.emnlp-main.2531/) (Loghmani et al., EMNLP 2026)

• Hidden Configuration Effects on Fairness in LLM-Based Multi-Agent Decision Systems (Loghmani et al., EMNLP 2026)

ACL

• Hamed Loghmani, Ninareh Mehrabi, Homa Hosseinmardi, and Amir Ghasemian. 2026. Hidden Configuration Effects on Fairness in LLM-Based Multi-Agent Decision Systems. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 57565–57589, Budapest, Hungary. Association for Computational Linguistics.