JALMBench: Benchmarking Jailbreak Vulnerabilities in Audio Language Models
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866912933132369920 |
|---|---|
| author | Peng, Zifan Liu, Yule Sun, Zhen Li, Mingchen Luo, Zeren Zheng, Jingyi Dong, Wenhan He, Xinlei Wang, Xuechao Xue, Yingjie Xu, Shengmin Huang, Xinyi |
| author_facet | Peng, Zifan Liu, Yule Sun, Zhen Li, Mingchen Luo, Zeren Zheng, Jingyi Dong, Wenhan He, Xinlei Wang, Xuechao Xue, Yingjie Xu, Shengmin Huang, Xinyi |
| contents | Large Audio Language Models (LALMs) have made significant progress. While increasingly deployed in real-world applications, LALMs face growing safety risks from jailbreak attacks that bypass safety alignment. However, there remains a lack of an adversarial audio dataset and a unified framework specifically designed to evaluate and compare jailbreak attacks against them. To address this gap, we introduce JALMBench, a comprehensive benchmark that assesses LALM safety against jailbreak attacks, comprising 11,316 text samples and 245,355 audio samples (>1,000 hours). JALMBench supports 12 mainstream LALMs, 8 attack methods (4 text-transferred and 4 audio-originated), and 5 defenses. We conduct in-depth analysis on attack efficiency, topic sensitivity, voice diversity, and model architecture. Additionally, we explore mitigation strategies for the attacks at both the prompt and response levels. Our systematic evaluation reveals that LALMs' safety is strongly influenced by modality and architectural choices: text-based safety alignment can partially transfer to audio inputs, and interleaved audio-text strategies enable more robust cross-modal generalization. Existing general-purpose moderation methods only slightly improve security, highlighting the need for defense methods specifically designed for LALMs. We hope our work can shed light on the design principles for building more robust LALMs. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_17568 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | JALMBench: Benchmarking Jailbreak Vulnerabilities in Audio Language Models Peng, Zifan Liu, Yule Sun, Zhen Li, Mingchen Luo, Zeren Zheng, Jingyi Dong, Wenhan He, Xinlei Wang, Xuechao Xue, Yingjie Xu, Shengmin Huang, Xinyi Cryptography and Security Artificial Intelligence Sound Audio and Speech Processing Large Audio Language Models (LALMs) have made significant progress. While increasingly deployed in real-world applications, LALMs face growing safety risks from jailbreak attacks that bypass safety alignment. However, there remains a lack of an adversarial audio dataset and a unified framework specifically designed to evaluate and compare jailbreak attacks against them. To address this gap, we introduce JALMBench, a comprehensive benchmark that assesses LALM safety against jailbreak attacks, comprising 11,316 text samples and 245,355 audio samples (>1,000 hours). JALMBench supports 12 mainstream LALMs, 8 attack methods (4 text-transferred and 4 audio-originated), and 5 defenses. We conduct in-depth analysis on attack efficiency, topic sensitivity, voice diversity, and model architecture. Additionally, we explore mitigation strategies for the attacks at both the prompt and response levels. Our systematic evaluation reveals that LALMs' safety is strongly influenced by modality and architectural choices: text-based safety alignment can partially transfer to audio inputs, and interleaved audio-text strategies enable more robust cross-modal generalization. Existing general-purpose moderation methods only slightly improve security, highlighting the need for defense methods specifically designed for LALMs. We hope our work can shed light on the design principles for building more robust LALMs. |
| title | JALMBench: Benchmarking Jailbreak Vulnerabilities in Audio Language Models |
| topic | Cryptography and Security Artificial Intelligence Sound Audio and Speech Processing |
| url | https://arxiv.org/abs/2505.17568 |