<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="son-etal-2026-unified">
    <titleInfo>
        <title>A Unified Mechanistic Analysis of Knowledge- and Safety-Based Refusals</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Yuri</namePart>
        <namePart type="family">Son</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Seunghee</namePart>
        <namePart type="family">Kim</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Hyuhng</namePart>
        <namePart type="given">Joon</namePart>
        <namePart type="family">Kim</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Taeuk</namePart>
        <namePart type="family">Kim</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) are increasingly trained to decline queries that fall outside their knowledge (knowledge-based refusal, KR) or violate safety policies (safety-based refusal, SR). Although KR and SR result in superficially similar responses, they have largely been studied in isolation, leaving open whether they share an underlying mechanism. We address this gap with a systematic study on a new dataset of 213 contrastive quadruples that jointly probe both refusal types. We find that KR and SR are governed by overlapping yet distinguishable mechanisms. Both share a refusal direction, yet the overlap is asymmetric: SR signals transfer more strongly to KR than the reverse. Type-specific specialization emerges mainly in upper layers, with KR aligning with uncertainty- and knowledge-related representations and SR with safety- and policy-related ones. We thus characterize refusal as a commit-then-specify process: a shared initial mechanism commits to refusing, then type-specific features in later layers specify whether the grounds are epistemic or normative.</abstract>
    <identifier type="citekey">son-etal-2026-unified</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.1111/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>25430</start>
            <end>25459</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T A Unified Mechanistic Analysis of Knowledge- and Safety-Based Refusals
%A Son, Yuri
%A Kim, Seunghee
%A Kim, Hyuhng Joon
%A Kim, Taeuk
%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 son-etal-2026-unified
%X Large language models (LLMs) are increasingly trained to decline queries that fall outside their knowledge (knowledge-based refusal, KR) or violate safety policies (safety-based refusal, SR). Although KR and SR result in superficially similar responses, they have largely been studied in isolation, leaving open whether they share an underlying mechanism. We address this gap with a systematic study on a new dataset of 213 contrastive quadruples that jointly probe both refusal types. We find that KR and SR are governed by overlapping yet distinguishable mechanisms. Both share a refusal direction, yet the overlap is asymmetric: SR signals transfer more strongly to KR than the reverse. Type-specific specialization emerges mainly in upper layers, with KR aligning with uncertainty- and knowledge-related representations and SR with safety- and policy-related ones. We thus characterize refusal as a commit-then-specify process: a shared initial mechanism commits to refusing, then type-specific features in later layers specify whether the grounds are epistemic or normative.
%U https://aclanthology.org/2026.emnlp-main.1111/
%P 25430-25459

Markdown (Informal)

[A Unified Mechanistic Analysis of Knowledge- and Safety-Based Refusals](https://aclanthology.org/2026.emnlp-main.1111/) (Son et al., EMNLP 2026)

• A Unified Mechanistic Analysis of Knowledge- and Safety-Based Refusals (Son et al., EMNLP 2026)

ACL

• Yuri Son, Seunghee Kim, Hyuhng Joon Kim, and Taeuk Kim. 2026. A Unified Mechanistic Analysis of Knowledge- and Safety-Based Refusals. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 25430–25459, Budapest, Hungary. Association for Computational Linguistics.