RL-JACK: Reinforcement Learning-powered Black-box Jailbreaking Attack against LLMs
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_ | 1866916285115269120 |
|---|---|
| author | Chen, Xuan Nie, Yuzhou Yan, Lu Mao, Yunshu Guo, Wenbo Zhang, Xiangyu |
| author_facet | Chen, Xuan Nie, Yuzhou Yan, Lu Mao, Yunshu Guo, Wenbo Zhang, Xiangyu |
| contents | Modern large language model (LLM) developers typically conduct a safety alignment to prevent an LLM from generating unethical or harmful content. Recent studies have discovered that the safety alignment of LLMs can be bypassed by jailbreaking prompts. These prompts are designed to create specific conversation scenarios with a harmful question embedded. Querying an LLM with such prompts can mislead the model into responding to the harmful question. The stochastic and random nature of existing genetic methods largely limits the effectiveness and efficiency of state-of-the-art (SOTA) jailbreaking attacks. In this paper, we propose RL-JACK, a novel black-box jailbreaking attack powered by deep reinforcement learning (DRL). We formulate the generation of jailbreaking prompts as a search problem and design a novel RL approach to solve it. Our method includes a series of customized designs to enhance the RL agent's learning efficiency in the jailbreaking context. Notably, we devise an LLM-facilitated action space that enables diverse action variations while constraining the overall search space. We propose a novel reward function that provides meaningful dense rewards for the agent toward achieving successful jailbreaking. Through extensive evaluations, we demonstrate that RL-JACK is overall much more effective than existing jailbreaking attacks against six SOTA LLMs, including large open-source models and commercial models. We also show the RL-JACK's resiliency against three SOTA defenses and its transferability across different models. Finally, we validate the insensitivity of RL-JACK to the variations in key hyper-parameters. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2406_08725 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | RL-JACK: Reinforcement Learning-powered Black-box Jailbreaking Attack against LLMs Chen, Xuan Nie, Yuzhou Yan, Lu Mao, Yunshu Guo, Wenbo Zhang, Xiangyu Cryptography and Security Modern large language model (LLM) developers typically conduct a safety alignment to prevent an LLM from generating unethical or harmful content. Recent studies have discovered that the safety alignment of LLMs can be bypassed by jailbreaking prompts. These prompts are designed to create specific conversation scenarios with a harmful question embedded. Querying an LLM with such prompts can mislead the model into responding to the harmful question. The stochastic and random nature of existing genetic methods largely limits the effectiveness and efficiency of state-of-the-art (SOTA) jailbreaking attacks. In this paper, we propose RL-JACK, a novel black-box jailbreaking attack powered by deep reinforcement learning (DRL). We formulate the generation of jailbreaking prompts as a search problem and design a novel RL approach to solve it. Our method includes a series of customized designs to enhance the RL agent's learning efficiency in the jailbreaking context. Notably, we devise an LLM-facilitated action space that enables diverse action variations while constraining the overall search space. We propose a novel reward function that provides meaningful dense rewards for the agent toward achieving successful jailbreaking. Through extensive evaluations, we demonstrate that RL-JACK is overall much more effective than existing jailbreaking attacks against six SOTA LLMs, including large open-source models and commercial models. We also show the RL-JACK's resiliency against three SOTA defenses and its transferability across different models. Finally, we validate the insensitivity of RL-JACK to the variations in key hyper-parameters. |
| title | RL-JACK: Reinforcement Learning-powered Black-box Jailbreaking Attack against LLMs |
| topic | Cryptography and Security |
| url | https://arxiv.org/abs/2406.08725 |