<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="su-etal-2026-persona">
    <titleInfo>
        <title>The Persona-Switch Jailbreak Backdoor in LLMs: From Payload Retrieval to Trigger-Gated Persona-State Selection</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Yanghao</namePart>
        <namePart type="family">Su</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Wenbo</namePart>
        <namePart type="family">Zhou</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Tianwei</namePart>
        <namePart type="family">Zhang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Han</namePart>
        <namePart type="family">Qiu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Weiming</namePart>
        <namePart type="family">Zhang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Nenghai</namePart>
        <namePart type="family">Yu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Jie</namePart>
        <namePart type="family">Zhang</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>Jailbreak backdoors typically pair a trigger with non-refusing responses, but under capability-based evaluation such payload-retrieval attacks often suppress refusal without producing actionable harm. We propose the persona-switch backdoor, where the trigger selects a persona-conditioned behavioral state rather than retrieving a fixed harmful response: style-consistent fine-tuning installs a persona, trigger training converts it into a dormant, trigger-addressable state, and persona-aligned framing amplifies harmful generation after activation. Across two model families and three personas, our attack substantially outperforms payload-retrieval baselines under capability-based evaluation while preserving high refusal rates on non-triggered inputs. The effect holds even for personas whose fine-tuning data contain no explicit harmful instructions, although they encode misaligned dispositions. Applying the same framing to clean and untriggered backdoored models yields near-zero harmful generation, showing that framing alone does not account for the attack and that harm depends on trigger activation of the poisoned persona-conditioned state. Activation steering supports a causal role for the persona-associated direction, pointing to representation-level intervention as a defense.</abstract>
    <identifier type="citekey">su-etal-2026-persona</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.1553/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>35434</start>
            <end>35449</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T The Persona-Switch Jailbreak Backdoor in LLMs: From Payload Retrieval to Trigger-Gated Persona-State Selection
%A Su, Yanghao
%A Zhou, Wenbo
%A Zhang, Tianwei
%A Qiu, Han
%A Zhang, Weiming
%A Yu, Nenghai
%A Zhang, Jie
%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 su-etal-2026-persona
%X Jailbreak backdoors typically pair a trigger with non-refusing responses, but under capability-based evaluation such payload-retrieval attacks often suppress refusal without producing actionable harm. We propose the persona-switch backdoor, where the trigger selects a persona-conditioned behavioral state rather than retrieving a fixed harmful response: style-consistent fine-tuning installs a persona, trigger training converts it into a dormant, trigger-addressable state, and persona-aligned framing amplifies harmful generation after activation. Across two model families and three personas, our attack substantially outperforms payload-retrieval baselines under capability-based evaluation while preserving high refusal rates on non-triggered inputs. The effect holds even for personas whose fine-tuning data contain no explicit harmful instructions, although they encode misaligned dispositions. Applying the same framing to clean and untriggered backdoored models yields near-zero harmful generation, showing that framing alone does not account for the attack and that harm depends on trigger activation of the poisoned persona-conditioned state. Activation steering supports a causal role for the persona-associated direction, pointing to representation-level intervention as a defense.
%U https://aclanthology.org/2026.emnlp-main.1553/
%P 35434-35449

Markdown (Informal)

[The Persona-Switch Jailbreak Backdoor in LLMs: From Payload Retrieval to Trigger-Gated Persona-State Selection](https://aclanthology.org/2026.emnlp-main.1553/) (Su et al., EMNLP 2026)

• The Persona-Switch Jailbreak Backdoor in LLMs: From Payload Retrieval to Trigger-Gated Persona-State Selection (Su et al., EMNLP 2026)

ACL

• Yanghao Su, Wenbo Zhou, Tianwei Zhang, Han Qiu, Weiming Zhang, Nenghai Yu, and Jie Zhang. 2026. The Persona-Switch Jailbreak Backdoor in LLMs: From Payload Retrieval to Trigger-Gated Persona-State Selection. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 35434–35449, Budapest, Hungary. Association for Computational Linguistics.