@inproceedings{zhengxian-etal-2026-slip,
title = "{SLIP}: Soft Label Mechanism and Key-Extraction-Guided {C}o{T}-based Defense Against Instruction Backdoor in {API}s",
author = "Zhengxian, Wu and
Wen, Juan and
Peng, Wanli and
Chang, Haowei and
Zhou, Yinghan and
Yiming, Xue",
editor = "Liakata, Maria and
Moreira, Viviane P. and
Zhang, Jiajun and
Jurgens, David",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
month = jul,
year = "2026",
address = "San Diego, California, United States",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-acl.896/",
pages = "18021--18042",
ISBN = "979-8-89176-395-1",
abstract = "Customized Large Language Model (LLM) agents face a critical security threat from black-box instruction backdoors, where malicious behaviors are covertly injected through hidden system instructions. Although existing prompt-based defenses can often detect poisoned inputs, they generally fail to recover correct outputs once the backdoor is activated. In this paper, we first conduct a mechanistic analysis of LLM behavior under instruction backdoors and reveal two pivotal phenomena: (1) cognitive override, in which backdoor triggers dominate the reasoning process and suppress task-relevant context, and (2) abnormal semantic correlation, where triggers establish excessively strong semantic associations with attacker-specified target labels. Based on these insights, we propose a $\textbf{S}$oft $\textbf{L}$abel mechanism and key-extraction-guided CoT-based defense against $\textbf{I}$nstruction backdoors in $\textbf{A}$PIs (SLIP). To counteract the cognitive override, the key-extraction-guided Chain-of-Thought (KCOT) explicitly guides the model to extract task-relevant keywords and phrases rather than only considering the single trigger or overall text semantics. To neutralize the trigger{'}s abnormal semantic correlation, the soft label mechanism (SLM) quantifies semantic correlations and employs statistical clustering to filter anomalous phrases before aggregating reliable keywords and phrases for prediction. Extensive experiments show that SLIP reduces the average attack success rate to 25.13{\%}, improves clean accuracy to 87.15{\%}, and outperforms state-of-the-art black-box defenses."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhengxian-etal-2026-slip">
<titleInfo>
<title>SLIP: Soft Label Mechanism and Key-Extraction-Guided CoT-based Defense Against Instruction Backdoor in APIs</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wu</namePart>
<namePart type="family">Zhengxian</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Juan</namePart>
<namePart type="family">Wen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wanli</namePart>
<namePart type="family">Peng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Haowei</namePart>
<namePart type="family">Chang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yinghan</namePart>
<namePart type="family">Zhou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xue</namePart>
<namePart type="family">Yiming</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: ACL 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Maria</namePart>
<namePart type="family">Liakata</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Viviane</namePart>
<namePart type="given">P</namePart>
<namePart type="family">Moreira</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiajun</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">David</namePart>
<namePart type="family">Jurgens</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, California, United States</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-395-1</identifier>
</relatedItem>
<abstract>Customized Large Language Model (LLM) agents face a critical security threat from black-box instruction backdoors, where malicious behaviors are covertly injected through hidden system instructions. Although existing prompt-based defenses can often detect poisoned inputs, they generally fail to recover correct outputs once the backdoor is activated. In this paper, we first conduct a mechanistic analysis of LLM behavior under instruction backdoors and reveal two pivotal phenomena: (1) cognitive override, in which backdoor triggers dominate the reasoning process and suppress task-relevant context, and (2) abnormal semantic correlation, where triggers establish excessively strong semantic associations with attacker-specified target labels. Based on these insights, we propose a Soft Label mechanism and key-extraction-guided CoT-based defense against Instruction backdoors in APIs (SLIP). To counteract the cognitive override, the key-extraction-guided Chain-of-Thought (KCOT) explicitly guides the model to extract task-relevant keywords and phrases rather than only considering the single trigger or overall text semantics. To neutralize the trigger’s abnormal semantic correlation, the soft label mechanism (SLM) quantifies semantic correlations and employs statistical clustering to filter anomalous phrases before aggregating reliable keywords and phrases for prediction. Extensive experiments show that SLIP reduces the average attack success rate to 25.13%, improves clean accuracy to 87.15%, and outperforms state-of-the-art black-box defenses.</abstract>
<identifier type="citekey">zhengxian-etal-2026-slip</identifier>
<location>
<url>https://aclanthology.org/2026.findings-acl.896/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>18021</start>
<end>18042</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T SLIP: Soft Label Mechanism and Key-Extraction-Guided CoT-based Defense Against Instruction Backdoor in APIs
%A Zhengxian, Wu
%A Wen, Juan
%A Peng, Wanli
%A Chang, Haowei
%A Zhou, Yinghan
%A Yiming, Xue
%Y Liakata, Maria
%Y Moreira, Viviane P.
%Y Zhang, Jiajun
%Y Jurgens, David
%S Findings of the Association for Computational Linguistics: ACL 2026
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, United States
%@ 979-8-89176-395-1
%F zhengxian-etal-2026-slip
%X Customized Large Language Model (LLM) agents face a critical security threat from black-box instruction backdoors, where malicious behaviors are covertly injected through hidden system instructions. Although existing prompt-based defenses can often detect poisoned inputs, they generally fail to recover correct outputs once the backdoor is activated. In this paper, we first conduct a mechanistic analysis of LLM behavior under instruction backdoors and reveal two pivotal phenomena: (1) cognitive override, in which backdoor triggers dominate the reasoning process and suppress task-relevant context, and (2) abnormal semantic correlation, where triggers establish excessively strong semantic associations with attacker-specified target labels. Based on these insights, we propose a Soft Label mechanism and key-extraction-guided CoT-based defense against Instruction backdoors in APIs (SLIP). To counteract the cognitive override, the key-extraction-guided Chain-of-Thought (KCOT) explicitly guides the model to extract task-relevant keywords and phrases rather than only considering the single trigger or overall text semantics. To neutralize the trigger’s abnormal semantic correlation, the soft label mechanism (SLM) quantifies semantic correlations and employs statistical clustering to filter anomalous phrases before aggregating reliable keywords and phrases for prediction. Extensive experiments show that SLIP reduces the average attack success rate to 25.13%, improves clean accuracy to 87.15%, and outperforms state-of-the-art black-box defenses.
%U https://aclanthology.org/2026.findings-acl.896/
%P 18021-18042
Markdown (Informal)
[SLIP: Soft Label Mechanism and Key-Extraction-Guided CoT-based Defense Against Instruction Backdoor in APIs](https://aclanthology.org/2026.findings-acl.896/) (Zhengxian et al., Findings 2026)
ACL