BadToken: Token-level Backdoor Attacks to Multi-modal Large Language Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Yuan, Zenghui, Shi, Jiawen, Zhou, Pan, Gong, Neil Zhenqiang, Sun, Lichao
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910886250151936
author Yuan, Zenghui
Shi, Jiawen
Zhou, Pan
Gong, Neil Zhenqiang
Sun, Lichao
author_facet Yuan, Zenghui
Shi, Jiawen
Zhou, Pan
Gong, Neil Zhenqiang
Sun, Lichao
contents Multi-modal large language models (MLLMs) extend large language models (LLMs) to process multi-modal information, enabling them to generate responses to image-text inputs. MLLMs have been incorporated into diverse multi-modal applications, such as autonomous driving and medical diagnosis, via plug-and-play without fine-tuning. This deployment paradigm increases the vulnerability of MLLMs to backdoor attacks. However, existing backdoor attacks against MLLMs achieve limited effectiveness and stealthiness. In this work, we propose BadToken, the first token-level backdoor attack to MLLMs. BadToken introduces two novel backdoor behaviors: Token-substitution and Token-addition, which enable flexible and stealthy attacks by making token-level modifications to the original output for backdoored inputs. We formulate a general optimization problem that considers the two backdoor behaviors to maximize the attack effectiveness. We evaluate BadToken on two open-source MLLMs and various tasks. Our results show that our attack maintains the model's utility while achieving high attack success rates and stealthiness. We also show the real-world threats of BadToken in two scenarios, i.e., autonomous driving and medical diagnosis. Furthermore, we consider defenses including fine-tuning and input purification. Our results highlight the threat of our attack.
format Preprint
id arxiv_https___arxiv_org_abs_2503_16023
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle BadToken: Token-level Backdoor Attacks to Multi-modal Large Language Models
Yuan, Zenghui
Shi, Jiawen
Zhou, Pan
Gong, Neil Zhenqiang
Sun, Lichao
Cryptography and Security
Multi-modal large language models (MLLMs) extend large language models (LLMs) to process multi-modal information, enabling them to generate responses to image-text inputs. MLLMs have been incorporated into diverse multi-modal applications, such as autonomous driving and medical diagnosis, via plug-and-play without fine-tuning. This deployment paradigm increases the vulnerability of MLLMs to backdoor attacks. However, existing backdoor attacks against MLLMs achieve limited effectiveness and stealthiness. In this work, we propose BadToken, the first token-level backdoor attack to MLLMs. BadToken introduces two novel backdoor behaviors: Token-substitution and Token-addition, which enable flexible and stealthy attacks by making token-level modifications to the original output for backdoored inputs. We formulate a general optimization problem that considers the two backdoor behaviors to maximize the attack effectiveness. We evaluate BadToken on two open-source MLLMs and various tasks. Our results show that our attack maintains the model's utility while achieving high attack success rates and stealthiness. We also show the real-world threats of BadToken in two scenarios, i.e., autonomous driving and medical diagnosis. Furthermore, we consider defenses including fine-tuning and input purification. Our results highlight the threat of our attack.
title BadToken: Token-level Backdoor Attacks to Multi-modal Large Language Models
topic Cryptography and Security
url https://arxiv.org/abs/2503.16023