ToxicTextCLIP: Text-Based Poisoning and Backdoor Attacks on CLIP Pre-training

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Yao, Xin, Zhao, Haiyang, Chen, Yimin, Guo, Jiawei, Huang, Kecheng, Zhao, Ming
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866908624415096832
author Yao, Xin
Zhao, Haiyang
Chen, Yimin
Guo, Jiawei
Huang, Kecheng
Zhao, Ming
author_facet Yao, Xin
Zhao, Haiyang
Chen, Yimin
Guo, Jiawei
Huang, Kecheng
Zhao, Ming
contents The Contrastive Language-Image Pretraining (CLIP) model has significantly advanced vision-language modeling by aligning image-text pairs from large-scale web data through self-supervised contrastive learning. Yet, its reliance on uncurated Internet-sourced data exposes it to data poisoning and backdoor risks. While existing studies primarily investigate image-based attacks, the text modality, which is equally central to CLIP's training, remains underexplored. In this work, we introduce ToxicTextCLIP, a framework for generating high-quality adversarial texts that target CLIP during the pre-training phase. The framework addresses two key challenges: semantic misalignment caused by background inconsistency with the target class, and the scarcity of background-consistent texts. To this end, ToxicTextCLIP iteratively applies: 1) a background-aware selector that prioritizes texts with background content aligned to the target class, and 2) a background-driven augmenter that generates semantically coherent and diverse poisoned samples. Extensive experiments on classification and retrieval tasks show that ToxicTextCLIP achieves up to 95.83% poisoning success and 98.68% backdoor Hit@1, while bypassing RoCLIP, CleanCLIP and SafeCLIP defenses. The source code can be accessed via https://github.com/xinyaocse/ToxicTextCLIP/.
format Preprint
id arxiv_https___arxiv_org_abs_2511_00446
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle ToxicTextCLIP: Text-Based Poisoning and Backdoor Attacks on CLIP Pre-training
Yao, Xin
Zhao, Haiyang
Chen, Yimin
Guo, Jiawei
Huang, Kecheng
Zhao, Ming
Computer Vision and Pattern Recognition
Cryptography and Security
Machine Learning
The Contrastive Language-Image Pretraining (CLIP) model has significantly advanced vision-language modeling by aligning image-text pairs from large-scale web data through self-supervised contrastive learning. Yet, its reliance on uncurated Internet-sourced data exposes it to data poisoning and backdoor risks. While existing studies primarily investigate image-based attacks, the text modality, which is equally central to CLIP's training, remains underexplored. In this work, we introduce ToxicTextCLIP, a framework for generating high-quality adversarial texts that target CLIP during the pre-training phase. The framework addresses two key challenges: semantic misalignment caused by background inconsistency with the target class, and the scarcity of background-consistent texts. To this end, ToxicTextCLIP iteratively applies: 1) a background-aware selector that prioritizes texts with background content aligned to the target class, and 2) a background-driven augmenter that generates semantically coherent and diverse poisoned samples. Extensive experiments on classification and retrieval tasks show that ToxicTextCLIP achieves up to 95.83% poisoning success and 98.68% backdoor Hit@1, while bypassing RoCLIP, CleanCLIP and SafeCLIP defenses. The source code can be accessed via https://github.com/xinyaocse/ToxicTextCLIP/.
title ToxicTextCLIP: Text-Based Poisoning and Backdoor Attacks on CLIP Pre-training
topic Computer Vision and Pattern Recognition
Cryptography and Security
Machine Learning
url https://arxiv.org/abs/2511.00446