Guardado en:
| Autores principales: | , , , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2024
|
| Materias: | |
| Acceso en línea: | https://arxiv.org/abs/2405.09113 |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866912230141853696 |
|---|---|
| author | Hu, Kai Yu, Weichen Li, Yining Chen, Kai Yao, Tianjun Li, Xiang Liu, Wenhe Yu, Lijun Shen, Zhiqiang Fredrikson, Matt |
| author_facet | Hu, Kai Yu, Weichen Li, Yining Chen, Kai Yao, Tianjun Li, Xiang Liu, Wenhe Yu, Lijun Shen, Zhiqiang Fredrikson, Matt |
| contents | Recent research indicates that large language models (LLMs) are susceptible to jailbreaking attacks that can generate harmful content. This paper introduces a novel token-level attack method, Adaptive Dense-to-Sparse Constrained Optimization (ADC), which has been shown to successfully jailbreak multiple open-source LLMs. Drawing inspiration from the difficulties of discrete token optimization, our method relaxes the discrete jailbreak optimization into a continuous optimization process while gradually increasing the sparsity of the optimizing vectors. This technique effectively bridges the gap between discrete and continuous space optimization. Experimental results demonstrate that our method is more effective and efficient than state-of-the-art token-level methods. On Harmbench, our approach achieves the highest attack success rate on seven out of eight LLMs compared to the latest jailbreak methods. Trigger Warning: This paper contains model behavior that can be offensive in nature. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2405_09113 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Efficient LLM Jailbreak via Adaptive Dense-to-sparse Constrained Optimization Hu, Kai Yu, Weichen Li, Yining Chen, Kai Yao, Tianjun Li, Xiang Liu, Wenhe Yu, Lijun Shen, Zhiqiang Fredrikson, Matt Machine Learning Recent research indicates that large language models (LLMs) are susceptible to jailbreaking attacks that can generate harmful content. This paper introduces a novel token-level attack method, Adaptive Dense-to-Sparse Constrained Optimization (ADC), which has been shown to successfully jailbreak multiple open-source LLMs. Drawing inspiration from the difficulties of discrete token optimization, our method relaxes the discrete jailbreak optimization into a continuous optimization process while gradually increasing the sparsity of the optimizing vectors. This technique effectively bridges the gap between discrete and continuous space optimization. Experimental results demonstrate that our method is more effective and efficient than state-of-the-art token-level methods. On Harmbench, our approach achieves the highest attack success rate on seven out of eight LLMs compared to the latest jailbreak methods. Trigger Warning: This paper contains model behavior that can be offensive in nature. |
| title | Efficient LLM Jailbreak via Adaptive Dense-to-sparse Constrained Optimization |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2405.09113 |