Test-Time Backdoor Attacks on Multimodal Large Language Models

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Lu, Dong, Pang, Tianyu, Du, Chao, Liu, Qian, Yang, Xianjun, Lin, Min
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866910329094537216
author Lu, Dong
Pang, Tianyu
Du, Chao
Liu, Qian
Yang, Xianjun
Lin, Min
author_facet Lu, Dong
Pang, Tianyu
Du, Chao
Liu, Qian
Yang, Xianjun
Lin, Min
contents Backdoor attacks are commonly executed by contaminating training data, such that a trigger can activate predetermined harmful effects during the test phase. In this work, we present AnyDoor, a test-time backdoor attack against multimodal large language models (MLLMs), which involves injecting the backdoor into the textual modality using adversarial test images (sharing the same universal perturbation), without requiring access to or modification of the training data. AnyDoor employs similar techniques used in universal adversarial attacks, but distinguishes itself by its ability to decouple the timing of setup and activation of harmful effects. In our experiments, we validate the effectiveness of AnyDoor against popular MLLMs such as LLaVA-1.5, MiniGPT-4, InstructBLIP, and BLIP-2, as well as provide comprehensive ablation studies. Notably, because the backdoor is injected by a universal perturbation, AnyDoor can dynamically change its backdoor trigger prompts/harmful effects, exposing a new challenge for defending against backdoor attacks. Our project page is available at https://sail-sg.github.io/AnyDoor/.
format Preprint
id arxiv_https___arxiv_org_abs_2402_08577
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Test-Time Backdoor Attacks on Multimodal Large Language Models
Lu, Dong
Pang, Tianyu
Du, Chao
Liu, Qian
Yang, Xianjun
Lin, Min
Computation and Language
Cryptography and Security
Computer Vision and Pattern Recognition
Machine Learning
Multimedia
Backdoor attacks are commonly executed by contaminating training data, such that a trigger can activate predetermined harmful effects during the test phase. In this work, we present AnyDoor, a test-time backdoor attack against multimodal large language models (MLLMs), which involves injecting the backdoor into the textual modality using adversarial test images (sharing the same universal perturbation), without requiring access to or modification of the training data. AnyDoor employs similar techniques used in universal adversarial attacks, but distinguishes itself by its ability to decouple the timing of setup and activation of harmful effects. In our experiments, we validate the effectiveness of AnyDoor against popular MLLMs such as LLaVA-1.5, MiniGPT-4, InstructBLIP, and BLIP-2, as well as provide comprehensive ablation studies. Notably, because the backdoor is injected by a universal perturbation, AnyDoor can dynamically change its backdoor trigger prompts/harmful effects, exposing a new challenge for defending against backdoor attacks. Our project page is available at https://sail-sg.github.io/AnyDoor/.
title Test-Time Backdoor Attacks on Multimodal Large Language Models
topic Computation and Language
Cryptography and Security
Computer Vision and Pattern Recognition
Machine Learning
Multimedia
url https://arxiv.org/abs/2402.08577