<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="joad-etal-2026-refusal">
    <titleInfo>
        <title>There Is More to Refusal in Large Language Models than a Single Direction</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Faaiz</namePart>
        <namePart type="family">Joad</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Majd</namePart>
        <namePart type="family">Hawasly</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Sabri</namePart>
        <namePart type="family">Boughorbel</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Nadir</namePart>
        <namePart type="family">Durrani</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Husrev</namePart>
        <namePart type="given">Taha</namePart>
        <namePart type="family">Sencar</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>Prior work argues that refusal in large language models is mediated by a single direction, enabling steering and abliteration. We show that this account is incomplete: across diverse refusal and non-compliance categories, refusal behaviors correspond to geometrically distinct directions in activation space. Yet activation steering along any refusal-related direction produces nearly identical refusal–over-refusal trade-offs, acting as a shared one-dimensional control knob. Thus, different directions primarily affect not whether the model refuses, but how it refuses. Using sparse autoencoders, we uncover a structured internal representation of refusal: a reusable core of shared refusal latents supplemented by style- and domain-specific latents. Linear interventions collapse this structure into uniform behavioral control, flattening mechanistic differences across refusal types. Our results reconcile the apparent simplicity of refusal steering with the diversity of refusal behaviors, and clarify the limits of linear interpretability for aligned model behavior.</abstract>
    <identifier type="citekey">joad-etal-2026-refusal</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.2606/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>59196</start>
            <end>59232</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T There Is More to Refusal in Large Language Models than a Single Direction
%A Joad, Faaiz
%A Hawasly, Majd
%A Boughorbel, Sabri
%A Durrani, Nadir
%A Sencar, Husrev Taha
%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 joad-etal-2026-refusal
%X Prior work argues that refusal in large language models is mediated by a single direction, enabling steering and abliteration. We show that this account is incomplete: across diverse refusal and non-compliance categories, refusal behaviors correspond to geometrically distinct directions in activation space. Yet activation steering along any refusal-related direction produces nearly identical refusal–over-refusal trade-offs, acting as a shared one-dimensional control knob. Thus, different directions primarily affect not whether the model refuses, but how it refuses. Using sparse autoencoders, we uncover a structured internal representation of refusal: a reusable core of shared refusal latents supplemented by style- and domain-specific latents. Linear interventions collapse this structure into uniform behavioral control, flattening mechanistic differences across refusal types. Our results reconcile the apparent simplicity of refusal steering with the diversity of refusal behaviors, and clarify the limits of linear interpretability for aligned model behavior.
%U https://aclanthology.org/2026.emnlp-main.2606/
%P 59196-59232

Markdown (Informal)

[There Is More to Refusal in Large Language Models than a Single Direction](https://aclanthology.org/2026.emnlp-main.2606/) (Joad et al., EMNLP 2026)

• There Is More to Refusal in Large Language Models than a Single Direction (Joad et al., EMNLP 2026)

ACL

• Faaiz Joad, Majd Hawasly, Sabri Boughorbel, Nadir Durrani, and Husrev Taha Sencar. 2026. There Is More to Refusal in Large Language Models than a Single Direction. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 59196–59232, Budapest, Hungary. Association for Computational Linguistics.