MEGen: Generative Backdoor into Large Language Models via Model Editing
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866909761578991616 |
|---|---|
| author | Qiu, Jiyang Ma, Xinbei Zhang, Zhuosheng Zhao, Hai Li, Yun Wang, Qianren |
| author_facet | Qiu, Jiyang Ma, Xinbei Zhang, Zhuosheng Zhao, Hai Li, Yun Wang, Qianren |
| contents | Large language models (LLMs) have exhibited remarkable versatility and adaptability, while their widespread adoption across various applications also raises critical safety concerns. This paper focuses on the impact of backdoored LLMs. Traditional backdoor injection methods are primarily limited to yes-or-no discriminative tasks, leading users to underestimate the potential risks of backdoored LLMs. Given the inherently generative nature of LLMs, this paper reveals that a generative backdoor injected into LLMs can expose the true safety risks in their applications. We propose an editing-based generative backdoor, named MEGen, aiming to expand the backdoor to generative tasks in a unified format of any text-to any text, leading to natural generations with a specific intention. Experiments show that MEGen achieves a high attack success rate by adjusting only a small set of local parameters with few-shot samples. Notably, we show that the backdoored model, when triggered, can freely output pre-set dangerous information while completing downstream tasks. Our work highlights that MEGen enables backdoors in LLMs to exhibit generative capabilities, causing potential safety risks by altering the generative style. The code is available at https://github.com/MonoQ-hub/MEGen. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2408_10722 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | MEGen: Generative Backdoor into Large Language Models via Model Editing Qiu, Jiyang Ma, Xinbei Zhang, Zhuosheng Zhao, Hai Li, Yun Wang, Qianren Computation and Language Artificial Intelligence Large language models (LLMs) have exhibited remarkable versatility and adaptability, while their widespread adoption across various applications also raises critical safety concerns. This paper focuses on the impact of backdoored LLMs. Traditional backdoor injection methods are primarily limited to yes-or-no discriminative tasks, leading users to underestimate the potential risks of backdoored LLMs. Given the inherently generative nature of LLMs, this paper reveals that a generative backdoor injected into LLMs can expose the true safety risks in their applications. We propose an editing-based generative backdoor, named MEGen, aiming to expand the backdoor to generative tasks in a unified format of any text-to any text, leading to natural generations with a specific intention. Experiments show that MEGen achieves a high attack success rate by adjusting only a small set of local parameters with few-shot samples. Notably, we show that the backdoored model, when triggered, can freely output pre-set dangerous information while completing downstream tasks. Our work highlights that MEGen enables backdoors in LLMs to exhibit generative capabilities, causing potential safety risks by altering the generative style. The code is available at https://github.com/MonoQ-hub/MEGen. |
| title | MEGen: Generative Backdoor into Large Language Models via Model Editing |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2408.10722 |