BadToken: Token-level Backdoor Attacks to Multi-modal Large Language Models
Preprint
Abstract
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.Keywords
All Related Versions
This publication has 12 references indexed in Scilit:
- Revisiting Backdoor Attacks against Large Vision-Language Models from Domain Shift2025
- TrojVLM: Backdoor Attack Against Vision Language Models2024
- Vision-language models for medical report generation and visual question answering: a reviewFrontiers in Artificial Intelligence, 2024
- DriveGPT4: Interpretable End-to-End Autonomous Driving Via Large Language ModelIEEE Robotics and Automation Letters, 2024
- Composite Backdoor Attacks Against Large Language Models2024
- Invisible Backdoor Attacks on Deep Neural Networks via Steganography and RegularizationIEEE Transactions on Dependable and Secure Computing, 2020
- Backdoor Embedding in Convolutional Neural Network Models via Invisible Perturbation2020
- ABS: Scanning Neural Networks for Back-doors by Artificial Brain Stimulation2019
- A New Backdoor Attack in CNNS by Training Set Corruption Without Label Poisoning2019
- DeepInspect: A Black-box Trojan Detection and Mitigation Framework for Deep Neural Networks2019