AudioJailbreak Benchmark

What is AudioJailbreak?

LAMs face jailbreak risks. AJailBench, our new benchmark, reveals leading LAMs lack robustness. Subtle audio perturbations significantly degrade their safety…

Released

2025-05-24

Modality

multimodal

Evaluates

多模态, Multimodal, 语言, Language, 安全, Safety, audio, jailbreak, LAM, 多模态模型, VLM, 语言理解, Comprehension, 安全对齐, Safety Alignment, 音频理解, Audio Understanding, 不支持, Unsupported

Openness

open

Importer

OpenCompass Hub

Reported scores

0

Source provenance

• Original evidence https://hub.opencompass.org.cn/dataset-detail/AudioJailbreak

AudioJailbreak paper, code and dataset

• Paper https://arxiv.org/abs/2505.15406
• Code https://github.com/mbzuai-nlp/AudioJailbreak
• Dataset https://huggingface.co/datasets/MBZUAI/AudioJailbreak
• Dataset https://huggingface.co/datasets/NEUQ-LIS-LAB/AudioJailbreak

Which sources cite AudioJailbreak?

• AudioJailbreak

Related 多模态, Multimodal, 语言, Language, 安全, Safety, audio, jailbreak, LAM, 多模态模型, VLM, 语言理解, Comprehension, 安全对齐, Safety Alignment, 音频理解, Audio Understanding, 不支持, Unsupported benchmarks

• 3DCodeBench
• A-Bench
• A-OKVQA
• AbsPyramid
• ActBench
• ActionAtlas
• ActiveFly-Bench
• Ada-LEval