<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhang-etal-2026-raset">
    <titleInfo>
        <title>RASET: Router-Agnostic Safety-Critical Expert Tuning Exposes Localized Safety Enforcement Failures in Mixture-of-Experts LLMs</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Zhibo</namePart>
        <namePart type="family">Zhang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Yuxi</namePart>
        <namePart type="family">Li</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Zhen</namePart>
        <namePart type="family">Ouyang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Ling</namePart>
        <namePart type="family">Shi</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Kailong</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) LLMs rely on sparse, router-driven expert activation, yet how safety alignment interacts with routed expert specialization remains underexplored. A common intuition is that safety behavior may be controlled by routing harmful requests to distinct refusal-oriented experts. In this work, we provide empirical evidence for a different picture: ***routing patterns in aligned MoE LLMs are largely topic-driven, while safety behavior can be altered with little change to the model’s intrinsic routing path***.\par Motivated by this observation, we present **RASET** (**R**outer-**A**gnostic **S**afety-critical **E**xpert **T**uning), a red-teaming framework that probes whether safety enforcement can be localized in a small subset of experts while preserving the model’s intrinsic routing behavior. **RASET** identifies safety-critical experts via a contrastive routing-sensitivity criterion and applies parameter-efficient tuning only to the selected experts, minimizing semantic disruption relative to router-steering interventions.\par Across five open-weight MoE backbones, **RASET** achieves high-quality safety-bypass yield (50.5% average ASR_hq, +37.6 points over the strongest baseline). These results reveal a distinct MoE safety risk, highlighting the need for expert-aware alignment, auditing, and robustness mechanisms in deployed MoE systems.</abstract>
    <identifier type="citekey">zhang-etal-2026-raset</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.2145/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>48770</start>
            <end>48789</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T RASET: Router-Agnostic Safety-Critical Expert Tuning Exposes Localized Safety Enforcement Failures in Mixture-of-Experts LLMs
%A Zhang, Zhibo
%A Li, Yuxi
%A Ouyang, Zhen
%A Shi, Ling
%A Wang, Kailong
%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 zhang-etal-2026-raset
%X Mixture-of-Experts (MoE) LLMs rely on sparse, router-driven expert activation, yet how safety alignment interacts with routed expert specialization remains underexplored. A common intuition is that safety behavior may be controlled by routing harmful requests to distinct refusal-oriented experts. In this work, we provide empirical evidence for a different picture: ***routing patterns in aligned MoE LLMs are largely topic-driven, while safety behavior can be altered with little change to the model’s intrinsic routing path***.\par Motivated by this observation, we present **RASET** (**R**outer-**A**gnostic **S**afety-critical **E**xpert **T**uning), a red-teaming framework that probes whether safety enforcement can be localized in a small subset of experts while preserving the model’s intrinsic routing behavior. **RASET** identifies safety-critical experts via a contrastive routing-sensitivity criterion and applies parameter-efficient tuning only to the selected experts, minimizing semantic disruption relative to router-steering interventions.\par Across five open-weight MoE backbones, **RASET** achieves high-quality safety-bypass yield (50.5% average ASR_hq, +37.6 points over the strongest baseline). These results reveal a distinct MoE safety risk, highlighting the need for expert-aware alignment, auditing, and robustness mechanisms in deployed MoE systems.
%U https://aclanthology.org/2026.emnlp-main.2145/
%P 48770-48789

Markdown (Informal)

[RASET: Router-Agnostic Safety-Critical Expert Tuning Exposes Localized Safety Enforcement Failures in Mixture-of-Experts LLMs](https://aclanthology.org/2026.emnlp-main.2145/) (Zhang et al., EMNLP 2026)

• RASET: Router-Agnostic Safety-Critical Expert Tuning Exposes Localized Safety Enforcement Failures in Mixture-of-Experts LLMs (Zhang et al., EMNLP 2026)

ACL

• Zhibo Zhang, Yuxi Li, Zhen Ouyang, Ling Shi, and Kailong Wang. 2026. RASET: Router-Agnostic Safety-Critical Expert Tuning Exposes Localized Safety Enforcement Failures in Mixture-of-Experts LLMs. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 48770–48789, Budapest, Hungary. Association for Computational Linguistics.