低频输入引发模型故障：LALMs中的安全风险研究

AI 摘要

研究人员提出了一种名为ILL的黑盒红队方法，利用听不见的低频波形暴露音频语言模型的漏洞，并设计了DRG防御机制来检测分布偏移并恢复准确性。实验表明，ILL可使模型准确率最多降低67个百分点，而DRG能将受攻击后的平均准确率从28.5%提升至46.1%。该研究揭示了LALMs中此前被忽视的安全风险。

正文节选

From Inaudible Inputs to Model Failures: Low-Frequency Safety Risks in LALMs Abstract Researchers propose a black-box red-teaming method using inaudible low-frequency waveforms to expose vulnerabilities in audio-language models, alongside a defense that detects distribution shifts and requests a second recording to recover accuracy. Large audio-language models (LALMs) have demonstrated strong capabilities in understanding diverse audio inputs. This diversity includes low-frequency signals that a

发布时间：—

抓取时间：2026-08-14 23:19

来源机构：Hugging Face