Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2411.12701 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916616515616768 |
|---|---|
| author | Ge, Huaizhi Li, Yiming Wang, Qifan Zhang, Yongfeng Tang, Ruixiang |
| author_facet | Ge, Huaizhi Li, Yiming Wang, Qifan Zhang, Yongfeng Tang, Ruixiang |
| contents | Large Language Models (LLMs) are known to be vulnerable to backdoor attacks, where triggers embedded in poisoned samples can maliciously alter LLMs' behaviors. In this paper, we move beyond attacking LLMs and instead examine backdoor attacks through the novel lens of natural language explanations. Specifically, we leverage LLMs' generative capabilities to produce human-readable explanations for their decisions, enabling direct comparisons between explanations for clean and poisoned samples. Our results show that backdoored models produce coherent explanations for clean inputs but diverse and logically flawed explanations for poisoned data, a pattern consistent across classification and generation tasks for different backdoor attacks. Further analysis reveals key insights into the explanation generation process. At the token level, explanation tokens associated with poisoned samples only appear in the final few transformer layers. At the sentence level, attention dynamics indicate that poisoned inputs shift attention away from the original input context during explanation generation. These findings enhance our understanding of backdoor mechanisms in LLMs and present a promising framework for detecting vulnerabilities through explainability. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2411_12701 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | When Backdoors Speak: Understanding LLM Backdoor Attacks Through Model-Generated Explanations Ge, Huaizhi Li, Yiming Wang, Qifan Zhang, Yongfeng Tang, Ruixiang Cryptography and Security Artificial Intelligence Large Language Models (LLMs) are known to be vulnerable to backdoor attacks, where triggers embedded in poisoned samples can maliciously alter LLMs' behaviors. In this paper, we move beyond attacking LLMs and instead examine backdoor attacks through the novel lens of natural language explanations. Specifically, we leverage LLMs' generative capabilities to produce human-readable explanations for their decisions, enabling direct comparisons between explanations for clean and poisoned samples. Our results show that backdoored models produce coherent explanations for clean inputs but diverse and logically flawed explanations for poisoned data, a pattern consistent across classification and generation tasks for different backdoor attacks. Further analysis reveals key insights into the explanation generation process. At the token level, explanation tokens associated with poisoned samples only appear in the final few transformer layers. At the sentence level, attention dynamics indicate that poisoned inputs shift attention away from the original input context during explanation generation. These findings enhance our understanding of backdoor mechanisms in LLMs and present a promising framework for detecting vulnerabilities through explainability. |
| title | When Backdoors Speak: Understanding LLM Backdoor Attacks Through Model-Generated Explanations |
| topic | Cryptography and Security Artificial Intelligence |
| url | https://arxiv.org/abs/2411.12701 |