<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="liang-etal-2026-routing">
    <titleInfo>
        <title>Routing-Aware Safety Alignment for Mixture-of-Experts Models</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Jiacheng</namePart>
        <namePart type="family">Liang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Yuhui</namePart>
        <namePart type="family">Wang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Tanqiu</namePart>
        <namePart type="family">Jiang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Ting</namePart>
        <namePart type="family">Wang</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>Mixture-of-Experts (MoE) language models introduce unique challenges for safety alignment due to their sparse routing mechanisms, which can enable degenerate optimization behaviors under standard full-parameter fine-tuning. In our preliminary experiments, we observe that naively applying full-parameter safety fine-tuning to MoE models can appear to improve safety through routing or expert dominance effects, rather than by directly repairing Safety-Critical Experts. To address this challenge, we propose RASA, a routing-aware expert-level alignment framework that explicitly repairs Safety-Critical Experts while preventing routing-based bypasses. RASA identifies experts disproportionately activated by successful jailbreaks, selectively fine-tunes only these experts under fixed routing, and subsequently enforces routing consistency with safety-aligned contexts. Across three representative MoE architectures spanning 1B to 47B total parameters, RASA achieves near-perfect single-turn robustness and cross-attack generalization, maintains low attack success rate under a RASA-aware adaptive attack, and extends effectively to held-out multi-turn jailbreaks, while preserving general capabilities and controlling over-refusal. Causal necessity and sufficiency tests confirm that the identified experts drive the safety repair. Our results suggest that robust MoE safety alignment benefits from targeted expert repair rather than global parameter updates, offering a practical and architecture-preserving alternative to prior approaches.</abstract>
    <identifier type="citekey">liang-etal-2026-routing</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.1047/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>24049</start>
            <end>24070</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Routing-Aware Safety Alignment for Mixture-of-Experts Models
%A Liang, Jiacheng
%A Wang, Yuhui
%A Jiang, Tanqiu
%A Wang, Ting
%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 liang-etal-2026-routing
%X Mixture-of-Experts (MoE) language models introduce unique challenges for safety alignment due to their sparse routing mechanisms, which can enable degenerate optimization behaviors under standard full-parameter fine-tuning. In our preliminary experiments, we observe that naively applying full-parameter safety fine-tuning to MoE models can appear to improve safety through routing or expert dominance effects, rather than by directly repairing Safety-Critical Experts. To address this challenge, we propose RASA, a routing-aware expert-level alignment framework that explicitly repairs Safety-Critical Experts while preventing routing-based bypasses. RASA identifies experts disproportionately activated by successful jailbreaks, selectively fine-tunes only these experts under fixed routing, and subsequently enforces routing consistency with safety-aligned contexts. Across three representative MoE architectures spanning 1B to 47B total parameters, RASA achieves near-perfect single-turn robustness and cross-attack generalization, maintains low attack success rate under a RASA-aware adaptive attack, and extends effectively to held-out multi-turn jailbreaks, while preserving general capabilities and controlling over-refusal. Causal necessity and sufficiency tests confirm that the identified experts drive the safety repair. Our results suggest that robust MoE safety alignment benefits from targeted expert repair rather than global parameter updates, offering a practical and architecture-preserving alternative to prior approaches.
%U https://aclanthology.org/2026.emnlp-main.1047/
%P 24049-24070

Markdown (Informal)

[Routing-Aware Safety Alignment for Mixture-of-Experts Models](https://aclanthology.org/2026.emnlp-main.1047/) (Liang et al., EMNLP 2026)

• Routing-Aware Safety Alignment for Mixture-of-Experts Models (Liang et al., EMNLP 2026)

ACL

• Jiacheng Liang, Yuhui Wang, Tanqiu Jiang, and Ting Wang. 2026. Routing-Aware Safety Alignment for Mixture-of-Experts Models. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 24049–24070, Budapest, Hungary. Association for Computational Linguistics.