CL-Attack: Textual Backdoor Attacks via Cross-Lingual Triggers
Fuente:
arXiv
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866908291350659072 |
|---|---|
| author | Zheng, Jingyi Hu, Tianyi Cong, Tianshuo He, Xinlei |
| author_facet | Zheng, Jingyi Hu, Tianyi Cong, Tianshuo He, Xinlei |
| contents | Backdoor attacks significantly compromise the security of large language models by triggering them to output specific and controlled content. Currently, triggers for textual backdoor attacks fall into two categories: fixed-token triggers and sentence-pattern triggers. However, the former are typically easy to identify and filter, while the latter, such as syntax and style, do not apply to all original samples and may lead to semantic shifts. In this paper, inspired by cross-lingual (CL) prompts of LLMs in real-world scenarios, we propose a higher-dimensional trigger method at the paragraph level, namely CL-attack. CL-attack injects the backdoor by using texts with specific structures that incorporate multiple languages, thereby offering greater stealthiness and universality compared to existing backdoor attack techniques. Extensive experiments on different tasks and model architectures demonstrate that CL-attack can achieve nearly 100% attack success rate with a low poisoning rate in both classification and generation tasks. We also empirically show that the CL-attack is more robust against current major defense methods compared to baseline backdoor attacks. Additionally, to mitigate CL-attack, we further develop a new defense called TranslateDefense, which can partially mitigate the impact of CL-attack. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2412_19037 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | CL-Attack: Textual Backdoor Attacks via Cross-Lingual Triggers Zheng, Jingyi Hu, Tianyi Cong, Tianshuo He, Xinlei Cryptography and Security Artificial Intelligence Backdoor attacks significantly compromise the security of large language models by triggering them to output specific and controlled content. Currently, triggers for textual backdoor attacks fall into two categories: fixed-token triggers and sentence-pattern triggers. However, the former are typically easy to identify and filter, while the latter, such as syntax and style, do not apply to all original samples and may lead to semantic shifts. In this paper, inspired by cross-lingual (CL) prompts of LLMs in real-world scenarios, we propose a higher-dimensional trigger method at the paragraph level, namely CL-attack. CL-attack injects the backdoor by using texts with specific structures that incorporate multiple languages, thereby offering greater stealthiness and universality compared to existing backdoor attack techniques. Extensive experiments on different tasks and model architectures demonstrate that CL-attack can achieve nearly 100% attack success rate with a low poisoning rate in both classification and generation tasks. We also empirically show that the CL-attack is more robust against current major defense methods compared to baseline backdoor attacks. Additionally, to mitigate CL-attack, we further develop a new defense called TranslateDefense, which can partially mitigate the impact of CL-attack. |
| title | CL-Attack: Textual Backdoor Attacks via Cross-Lingual Triggers |
| topic | Cryptography and Security Artificial Intelligence |
| url | https://arxiv.org/abs/2412.19037 |