@inproceedings{huang-etal-2024-composite,
title = "Composite Backdoor Attacks Against Large Language Models",
author = "Huang, Hai and
Zhao, Zhengyu and
Backes, Michael and
Shen, Yun and
Zhang, Yang",
editor = "Duh, Kevin and
Gomez, Helena and
Bethard, Steven",
booktitle = "Findings of the Association for Computational Linguistics: NAACL 2024",
month = jun,
year = "2024",
address = "Mexico City, Mexico",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2024.findings-naacl.94/",
doi = "10.18653/v1/2024.findings-naacl.94",
pages = "1459--1472",
abstract = "Large language models (LLMs) have demonstrated superior performance compared to previous methods on various tasks, and often serve as the foundation models for many researches and services. However, the untrustworthy third-party LLMs may covertly introduce vulnerabilities for downstream tasks. In this paper, we explore the vulnerability of LLMs through the lens of backdoor attacks. Different from existing backdoor attacks against LLMs, ours scatters multiple trigger keys in different prompt components. Such a Composite Backdoor Attack (CBA) is shown to be stealthier than implanting the same multiple trigger keys in only a single component. CBA ensures that the backdoor is activated only when all trigger keys appear. Our experiments demonstrate that CBA is effective in both natural language processing (NLP) and multimodal tasks. For instance, with 3{\%} poisoning samples against the LLaMA-7B model on the Emotion dataset, our attack achieves a 100{\%} Attack Success Rate (ASR) with a False Triggered Rate (FTR) below 2.06{\%} and negligible model accuracy degradation. Our work highlights the necessity of increased security research on the trustworthiness of foundation LLMs."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="huang-etal-2024-composite">
<titleInfo>
<title>Composite Backdoor Attacks Against Large Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hai</namePart>
<namePart type="family">Huang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhengyu</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Michael</namePart>
<namePart type="family">Backes</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yun</namePart>
<namePart type="family">Shen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yang</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2024-06</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: NAACL 2024</title>
</titleInfo>
<name type="personal">
<namePart type="given">Kevin</namePart>
<namePart type="family">Duh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Helena</namePart>
<namePart type="family">Gomez</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Steven</namePart>
<namePart type="family">Bethard</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Mexico City, Mexico</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Large language models (LLMs) have demonstrated superior performance compared to previous methods on various tasks, and often serve as the foundation models for many researches and services. However, the untrustworthy third-party LLMs may covertly introduce vulnerabilities for downstream tasks. In this paper, we explore the vulnerability of LLMs through the lens of backdoor attacks. Different from existing backdoor attacks against LLMs, ours scatters multiple trigger keys in different prompt components. Such a Composite Backdoor Attack (CBA) is shown to be stealthier than implanting the same multiple trigger keys in only a single component. CBA ensures that the backdoor is activated only when all trigger keys appear. Our experiments demonstrate that CBA is effective in both natural language processing (NLP) and multimodal tasks. For instance, with 3% poisoning samples against the LLaMA-7B model on the Emotion dataset, our attack achieves a 100% Attack Success Rate (ASR) with a False Triggered Rate (FTR) below 2.06% and negligible model accuracy degradation. Our work highlights the necessity of increased security research on the trustworthiness of foundation LLMs.</abstract>
<identifier type="citekey">huang-etal-2024-composite</identifier>
<identifier type="doi">10.18653/v1/2024.findings-naacl.94</identifier>
<location>
<url>https://aclanthology.org/2024.findings-naacl.94/</url>
</location>
<part>
<date>2024-06</date>
<extent unit="page">
<start>1459</start>
<end>1472</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Composite Backdoor Attacks Against Large Language Models
%A Huang, Hai
%A Zhao, Zhengyu
%A Backes, Michael
%A Shen, Yun
%A Zhang, Yang
%Y Duh, Kevin
%Y Gomez, Helena
%Y Bethard, Steven
%S Findings of the Association for Computational Linguistics: NAACL 2024
%D 2024
%8 June
%I Association for Computational Linguistics
%C Mexico City, Mexico
%F huang-etal-2024-composite
%X Large language models (LLMs) have demonstrated superior performance compared to previous methods on various tasks, and often serve as the foundation models for many researches and services. However, the untrustworthy third-party LLMs may covertly introduce vulnerabilities for downstream tasks. In this paper, we explore the vulnerability of LLMs through the lens of backdoor attacks. Different from existing backdoor attacks against LLMs, ours scatters multiple trigger keys in different prompt components. Such a Composite Backdoor Attack (CBA) is shown to be stealthier than implanting the same multiple trigger keys in only a single component. CBA ensures that the backdoor is activated only when all trigger keys appear. Our experiments demonstrate that CBA is effective in both natural language processing (NLP) and multimodal tasks. For instance, with 3% poisoning samples against the LLaMA-7B model on the Emotion dataset, our attack achieves a 100% Attack Success Rate (ASR) with a False Triggered Rate (FTR) below 2.06% and negligible model accuracy degradation. Our work highlights the necessity of increased security research on the trustworthiness of foundation LLMs.
%R 10.18653/v1/2024.findings-naacl.94
%U https://aclanthology.org/2024.findings-naacl.94/
%U https://doi.org/10.18653/v1/2024.findings-naacl.94
%P 1459-1472
Markdown (Informal)
[Composite Backdoor Attacks Against Large Language Models](https://aclanthology.org/2024.findings-naacl.94/) (Huang et al., Findings 2024)
ACL
- Hai Huang, Zhengyu Zhao, Michael Backes, Yun Shen, and Yang Zhang. 2024. Composite Backdoor Attacks Against Large Language Models. In Findings of the Association for Computational Linguistics: NAACL 2024, pages 1459–1472, Mexico City, Mexico. Association for Computational Linguistics.