@inproceedings{choi-etal-2026-easily,
title = "``{I}'' can easily jailbreak {LLM}s",
author = "Choi, Eunkyung and
Kim, Yeeun and
Kang, Juheon and
Hwang, Wonseok",
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.2450/",
pages = "52589--52608",
ISBN = "979-8-89176-462-0",
abstract = "Human-AI alignment is critical for ethical and safe use of large language models (LLMs). In this paper, we show that safety-aligned LLMs exhibit certain patterns in how refusals begin{--}such as disproportionately beginning responses with ``I''. Can these patterns be exploited to break safety alignment? Here, we introduce ACHILLES, a lightweight RLVR-based attack that uses pattern-based verifiers to exploit these refusal-correlated behaviors, achieving substantial increases in attack success rates across diverse LLM families and exceeding 90{\%} on safety-enhanced reasoning models, while preserving downstream capabilities. We further demonstrate that this vulnerability originates from domain-wise imbalances (safety vs non-safety) in safety alignment datasets, where safety responses share exploitable refusal-opening patterns. We additionally observe that safety-aligned models exhibit highly separable distributions between refusal and compliance responses in hidden representation space. To address this, we construct THETIS, a defense-oriented safety alignment dataset that reduces shallow refusal-pattern correlations. Models trained on THETIS maintain near-zero attack success rates against both ACHILLES and LLM-verifier based RLVR attacks, while largely preserving downstream performance."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="choi-etal-2026-easily">
<titleInfo>
<title>“I” can easily jailbreak LLMs</title>
</titleInfo>
<name type="personal">
<namePart type="given">Eunkyung</namePart>
<namePart type="family">Choi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yeeun</namePart>
<namePart type="family">Kim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Juheon</namePart>
<namePart type="family">Kang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wonseok</namePart>
<namePart type="family">Hwang</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>Human-AI alignment is critical for ethical and safe use of large language models (LLMs). In this paper, we show that safety-aligned LLMs exhibit certain patterns in how refusals begin–such as disproportionately beginning responses with “I”. Can these patterns be exploited to break safety alignment? Here, we introduce ACHILLES, a lightweight RLVR-based attack that uses pattern-based verifiers to exploit these refusal-correlated behaviors, achieving substantial increases in attack success rates across diverse LLM families and exceeding 90% on safety-enhanced reasoning models, while preserving downstream capabilities. We further demonstrate that this vulnerability originates from domain-wise imbalances (safety vs non-safety) in safety alignment datasets, where safety responses share exploitable refusal-opening patterns. We additionally observe that safety-aligned models exhibit highly separable distributions between refusal and compliance responses in hidden representation space. To address this, we construct THETIS, a defense-oriented safety alignment dataset that reduces shallow refusal-pattern correlations. Models trained on THETIS maintain near-zero attack success rates against both ACHILLES and LLM-verifier based RLVR attacks, while largely preserving downstream performance.</abstract>
<identifier type="citekey">choi-etal-2026-easily</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.2450/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>52589</start>
<end>52608</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T “I” can easily jailbreak LLMs
%A Choi, Eunkyung
%A Kim, Yeeun
%A Kang, Juheon
%A Hwang, Wonseok
%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 choi-etal-2026-easily
%X Human-AI alignment is critical for ethical and safe use of large language models (LLMs). In this paper, we show that safety-aligned LLMs exhibit certain patterns in how refusals begin–such as disproportionately beginning responses with “I”. Can these patterns be exploited to break safety alignment? Here, we introduce ACHILLES, a lightweight RLVR-based attack that uses pattern-based verifiers to exploit these refusal-correlated behaviors, achieving substantial increases in attack success rates across diverse LLM families and exceeding 90% on safety-enhanced reasoning models, while preserving downstream capabilities. We further demonstrate that this vulnerability originates from domain-wise imbalances (safety vs non-safety) in safety alignment datasets, where safety responses share exploitable refusal-opening patterns. We additionally observe that safety-aligned models exhibit highly separable distributions between refusal and compliance responses in hidden representation space. To address this, we construct THETIS, a defense-oriented safety alignment dataset that reduces shallow refusal-pattern correlations. Models trained on THETIS maintain near-zero attack success rates against both ACHILLES and LLM-verifier based RLVR attacks, while largely preserving downstream performance.
%U https://aclanthology.org/2026.findings-emnlp.2450/
%P 52589-52608
Markdown (Informal)
["I" can easily jailbreak LLMs](https://aclanthology.org/2026.findings-emnlp.2450/) (Choi et al., Findings 2026)
ACL
- Eunkyung Choi, Yeeun Kim, Juheon Kang, and Wonseok Hwang. 2026. "I" can easily jailbreak LLMs. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 52589–52608, Budapest, Hungary. Association for Computational Linguistics.