@inproceedings{rahman-etal-2026-attck,
title = "{ATTCK}-{R}: Knowledge-Aware Long-Horizon Jailbreaking for Protected {LLM}s",
author = "Rahman, Md Mushfiqur and
Rohde, Tobias and
Tripathi, Rahul and
Elkhalil, Khalil and
Lybarger, Kevin",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.2296/",
doi = "10.18653/2026.findings-emnlp.2296",
pages = "49229--49255",
ISBN = "979-8-89176-462-0",
abstract = "As large language models are deployed behind increasingly sophisticated safety mechanisms, evaluating their robustness under realistic adversarial conditions becomes increasingly difficult and critical. Existing jailbreaking evaluations often rely on static prompts, limited interaction horizons, or lightly protected systems, failing to reflect protected deployment settings with prompt-based safety protections. We show that the introduction of a prompt-based safety protection reduces the average attack success rate of a strong multi-turn baseline from 0.896 to 0.608 (32.1{\%} relative), demonstrating that unprotected evaluations substantially overestimate real-world vulnerabilities. We propose Adaptive Traceback and Tactic Curation for Knowledge-based Red-teaming (ATTCK-R), a planning-driven multi-turn red teaming framework designed for protected black-box systems. ATTCK-R models jailbreaking as a long-horizon trajectory optimization problem using a structured belief state, introduces optimization-based tactic curation to abstract reusable strategies from prior successes and failures, and employs informed traceback to resume attacks from strategically meaningful states rather than restarting. Evaluated across 9 widely used LLMs configured with a single prompt-based safety protection layer, ATTCK-R consistently outperforms strong recent multi-turn jailbreaking frameworks. Ablation results confirm that long-horizon planning and tactic curation improve attack effectiveness, while adaptive traceback improves efficiency through structured recovery from conversational failure."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="rahman-etal-2026-attck">
<titleInfo>
<title>ATTCK-R: Knowledge-Aware Long-Horizon Jailbreaking for Protected LLMs</title>
</titleInfo>
<name type="personal">
<namePart type="given">Md</namePart>
<namePart type="given">Mushfiqur</namePart>
<namePart type="family">Rahman</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tobias</namePart>
<namePart type="family">Rohde</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Rahul</namePart>
<namePart type="family">Tripathi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Khalil</namePart>
<namePart type="family">Elkhalil</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kevin</namePart>
<namePart type="family">Lybarger</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>As large language models are deployed behind increasingly sophisticated safety mechanisms, evaluating their robustness under realistic adversarial conditions becomes increasingly difficult and critical. Existing jailbreaking evaluations often rely on static prompts, limited interaction horizons, or lightly protected systems, failing to reflect protected deployment settings with prompt-based safety protections. We show that the introduction of a prompt-based safety protection reduces the average attack success rate of a strong multi-turn baseline from 0.896 to 0.608 (32.1% relative), demonstrating that unprotected evaluations substantially overestimate real-world vulnerabilities. We propose Adaptive Traceback and Tactic Curation for Knowledge-based Red-teaming (ATTCK-R), a planning-driven multi-turn red teaming framework designed for protected black-box systems. ATTCK-R models jailbreaking as a long-horizon trajectory optimization problem using a structured belief state, introduces optimization-based tactic curation to abstract reusable strategies from prior successes and failures, and employs informed traceback to resume attacks from strategically meaningful states rather than restarting. Evaluated across 9 widely used LLMs configured with a single prompt-based safety protection layer, ATTCK-R consistently outperforms strong recent multi-turn jailbreaking frameworks. Ablation results confirm that long-horizon planning and tactic curation improve attack effectiveness, while adaptive traceback improves efficiency through structured recovery from conversational failure.</abstract>
<identifier type="citekey">rahman-etal-2026-attck</identifier>
<identifier type="doi">10.18653/2026.findings-emnlp.2296</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.2296/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>49229</start>
<end>49255</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T ATTCK-R: Knowledge-Aware Long-Horizon Jailbreaking for Protected LLMs
%A Rahman, Md Mushfiqur
%A Rohde, Tobias
%A Tripathi, Rahul
%A Elkhalil, Khalil
%A Lybarger, Kevin
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F rahman-etal-2026-attck
%X As large language models are deployed behind increasingly sophisticated safety mechanisms, evaluating their robustness under realistic adversarial conditions becomes increasingly difficult and critical. Existing jailbreaking evaluations often rely on static prompts, limited interaction horizons, or lightly protected systems, failing to reflect protected deployment settings with prompt-based safety protections. We show that the introduction of a prompt-based safety protection reduces the average attack success rate of a strong multi-turn baseline from 0.896 to 0.608 (32.1% relative), demonstrating that unprotected evaluations substantially overestimate real-world vulnerabilities. We propose Adaptive Traceback and Tactic Curation for Knowledge-based Red-teaming (ATTCK-R), a planning-driven multi-turn red teaming framework designed for protected black-box systems. ATTCK-R models jailbreaking as a long-horizon trajectory optimization problem using a structured belief state, introduces optimization-based tactic curation to abstract reusable strategies from prior successes and failures, and employs informed traceback to resume attacks from strategically meaningful states rather than restarting. Evaluated across 9 widely used LLMs configured with a single prompt-based safety protection layer, ATTCK-R consistently outperforms strong recent multi-turn jailbreaking frameworks. Ablation results confirm that long-horizon planning and tactic curation improve attack effectiveness, while adaptive traceback improves efficiency through structured recovery from conversational failure.
%R 10.18653/2026.findings-emnlp.2296
%U https://aclanthology.org/2026.findings-emnlp.2296/
%U https://doi.org/10.18653/2026.findings-emnlp.2296
%P 49229-49255
Markdown (Informal)
[ATTCK-R: Knowledge-Aware Long-Horizon Jailbreaking for Protected LLMs](https://aclanthology.org/2026.findings-emnlp.2296/) (Rahman et al., Findings 2026)
ACL