<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="manjusha-emani-2026-transferring">
    <titleInfo>
        <title>Transferring Text Jailbreaks to Audio in Tool-Using LLM Agents</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Kashyap</namePart>
        <namePart type="given">R</namePart>
        <namePart type="family">Manjusha</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Murali</namePart>
        <namePart type="family">Emani</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2026-11</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Proceedings of the Workshop on Secure, Safe, and Privacy-Preserving Speech and Audio AI (TrustAudio 2026)</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Nancy</namePart>
            <namePart type="given">F</namePart>
            <namePart type="family">Chen</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Yiming</namePart>
            <namePart type="family">Chen</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Xiaoxue</namePart>
            <namePart type="family">Gao</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Haizhou</namePart>
            <namePart type="family">Li</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Björn</namePart>
            <namePart type="family">Schuller</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Berrak</namePart>
            <namePart type="family">Sisman</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Xianghu</namePart>
            <namePart type="family">Yue</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Hengqin, Zhuhai, China</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
        <identifier type="isbn">979-8-89176-452-1</identifier>
    </relatedItem>
    <abstract>Safety alignment in large audio-language models (LALMs) is typically evaluated through single-turn interactions with direct prompts. In real-world deployments, however, these models are often integrated into agent frameworks that introduce developer-controlled system prompts, tool-calling loops, and intermediate encoding steps between user speech and model input. This raises a practical question: can a text-based jailbreak remain effective after being converted to speech by an off-the-shelf text-to-speech (TTS) system and processed through an agent? We investigate this question using 270 prohibited questions spanning nine policy categories, an open-weight TTS model with 82 million parameters, and a minimal LangChain question-answering agent whose system prompt combines a persona with a fuzzing-derived template. Across three audio-capable models, we observe attack success rates (ASRs) ranging from 57.0% to 64.1%. To assess the impact of synthesis errors, we use automatic speech recognition to verify the generated audio and find that 87.4% of clips are transcribed exactly. Our analysis indicates that faithfully rendered speech alone achieves an ASR above 50% on every model tested, while manual evaluation of a 50-response sample agrees with the automatic judge in 92% of cases. These findings suggest that the agent framework does not compensate for the loss of safety robustness introduced by the text-to-audio transition. Even a readily available TTS pipeline requiring only minutes of computation can therefore provide a practical attack surface for transferring text-based jailbreaks to audio-agent settings.</abstract>
    <identifier type="citekey">manjusha-emani-2026-transferring</identifier>
    <location>
        <url>https://aclanthology.org/2026.trustaudio-1.3/</url>
    </location>
    <part>
        <date>2026-11</date>
        <extent unit="page">
            <start>24</start>
            <end>28</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Transferring Text Jailbreaks to Audio in Tool-Using LLM Agents
%A Manjusha, Kashyap R.
%A Emani, Murali
%Y Chen, Nancy F.
%Y Chen, Yiming
%Y Gao, Xiaoxue
%Y Li, Haizhou
%Y Schuller, Björn
%Y Sisman, Berrak
%Y Yue, Xianghu
%S Proceedings of the Workshop on Secure, Safe, and Privacy-Preserving Speech and Audio AI (TrustAudio 2026)
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, Zhuhai, China
%@ 979-8-89176-452-1
%F manjusha-emani-2026-transferring
%X Safety alignment in large audio-language models (LALMs) is typically evaluated through single-turn interactions with direct prompts. In real-world deployments, however, these models are often integrated into agent frameworks that introduce developer-controlled system prompts, tool-calling loops, and intermediate encoding steps between user speech and model input. This raises a practical question: can a text-based jailbreak remain effective after being converted to speech by an off-the-shelf text-to-speech (TTS) system and processed through an agent? We investigate this question using 270 prohibited questions spanning nine policy categories, an open-weight TTS model with 82 million parameters, and a minimal LangChain question-answering agent whose system prompt combines a persona with a fuzzing-derived template. Across three audio-capable models, we observe attack success rates (ASRs) ranging from 57.0% to 64.1%. To assess the impact of synthesis errors, we use automatic speech recognition to verify the generated audio and find that 87.4% of clips are transcribed exactly. Our analysis indicates that faithfully rendered speech alone achieves an ASR above 50% on every model tested, while manual evaluation of a 50-response sample agrees with the automatic judge in 92% of cases. These findings suggest that the agent framework does not compensate for the loss of safety robustness introduced by the text-to-audio transition. Even a readily available TTS pipeline requiring only minutes of computation can therefore provide a practical attack surface for transferring text-based jailbreaks to audio-agent settings.
%U https://aclanthology.org/2026.trustaudio-1.3/
%P 24-28

Markdown (Informal)

[Transferring Text Jailbreaks to Audio in Tool-Using LLM Agents](https://aclanthology.org/2026.trustaudio-1.3/) (Manjusha & Emani, TrustAudio 2026)

• Transferring Text Jailbreaks to Audio in Tool-Using LLM Agents (Manjusha & Emani, TrustAudio 2026)

ACL

• Kashyap R Manjusha and Murali Emani. 2026. Transferring Text Jailbreaks to Audio in Tool-Using LLM Agents. In Proceedings of the Workshop on Secure, Safe, and Privacy-Preserving Speech and Audio AI (TrustAudio 2026), pages 24–28, Hengqin, Zhuhai, China. Association for Computational Linguistics.