<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="huang-etal-2026-coda-mitigating">
    <titleInfo>
        <title>CODA: Mitigating Over-Refusal and Jailbreaks in Large Language Models via Codebook-Orthogonal Direction Alignment</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Feng</namePart>
        <namePart type="family">Huang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Yunqi</namePart>
        <namePart type="family">Zhang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Qiang</namePart>
        <namePart type="family">Huang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Qing</namePart>
        <namePart type="family">Yu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Yongfeng</namePart>
        <namePart type="family">Huang</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>Safety alignment plays a vital role in ensuring the reliability of Large Language Models (LLMs) by mitigating risks from malicious queries. To avoid the high computational costs of post-training, recent research has increasingly shifted toward inference-time alignment via continuous activation steering.\par However, the geometric entanglement of harmful and benign technical representations within these continuous spaces impedes precise calibration, inevitably trapping models in a severe dilemma between jailbreak vulnerability and benign over-refusal. To address this, we propose Codebook-Orthogonal Direction Alignment (CODA), an inference-time framework that projects hidden states onto a vector-quantized codebook to serve as a discrete control interface.\par Specifically, to break continuous directional entanglement, CODA estimates steering directions over discrete codebook anchors. To eliminate the geometric interference triggering over-refusals, CODA softly orthogonalizes the safety direction with respect to benign technical intent.\par Finally, to prevent rigid over-correction, CODA forwards the decoupled vectors to yield dynamically adjusted steering strengths during generation via risk gating and exponential time decay. Experiments across a diverse set of open-weight LLMs demonstrate that CODA consistently yields a stronger safety–utility trade-off than continuous steering baselines, indicating that discrete representations provide a more effective mechanism for disentangling safety from utility.</abstract>
    <identifier type="citekey">huang-etal-2026-coda-mitigating</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.1243/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>28326</start>
            <end>28348</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T CODA: Mitigating Over-Refusal and Jailbreaks in Large Language Models via Codebook-Orthogonal Direction Alignment
%A Huang, Feng
%A Zhang, Yunqi
%A Huang, Qiang
%A Yu, Qing
%A Huang, Yongfeng
%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 huang-etal-2026-coda-mitigating
%X Safety alignment plays a vital role in ensuring the reliability of Large Language Models (LLMs) by mitigating risks from malicious queries. To avoid the high computational costs of post-training, recent research has increasingly shifted toward inference-time alignment via continuous activation steering.\par However, the geometric entanglement of harmful and benign technical representations within these continuous spaces impedes precise calibration, inevitably trapping models in a severe dilemma between jailbreak vulnerability and benign over-refusal. To address this, we propose Codebook-Orthogonal Direction Alignment (CODA), an inference-time framework that projects hidden states onto a vector-quantized codebook to serve as a discrete control interface.\par Specifically, to break continuous directional entanglement, CODA estimates steering directions over discrete codebook anchors. To eliminate the geometric interference triggering over-refusals, CODA softly orthogonalizes the safety direction with respect to benign technical intent.\par Finally, to prevent rigid over-correction, CODA forwards the decoupled vectors to yield dynamically adjusted steering strengths during generation via risk gating and exponential time decay. Experiments across a diverse set of open-weight LLMs demonstrate that CODA consistently yields a stronger safety–utility trade-off than continuous steering baselines, indicating that discrete representations provide a more effective mechanism for disentangling safety from utility.
%U https://aclanthology.org/2026.emnlp-main.1243/
%P 28326-28348

Markdown (Informal)

[CODA: Mitigating Over-Refusal and Jailbreaks in Large Language Models via Codebook-Orthogonal Direction Alignment](https://aclanthology.org/2026.emnlp-main.1243/) (Huang et al., EMNLP 2026)

• CODA: Mitigating Over-Refusal and Jailbreaks in Large Language Models via Codebook-Orthogonal Direction Alignment (Huang et al., EMNLP 2026)

ACL

• Feng Huang, Yunqi Zhang, Qiang Huang, Qing Yu, and Yongfeng Huang. 2026. CODA: Mitigating Over-Refusal and Jailbreaks in Large Language Models via Codebook-Orthogonal Direction Alignment. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 28326–28348, Budapest, Hungary. Association for Computational Linguistics.