arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2508.03365cs.SDcs.AIcs.CReess.AS
收藏

当良好声音变得对抗性:利用无害输入对音频-语言模型进行劫持

When Good Sounds Go Adversarial: Jailbreaking Audio-Language Models with Benign Inputs

Yonsei University, Seoul, South Korea(延世大学);Seoul National University, Seoul, South Korea(首尔国立大学);Pohang University of Science(坡桑科学大学)

Hiskias Dingeto, Taeyoun Kwon, Dasol Choi, Bodam Kim, DongGeon Lee, Haon Park, JaeHoon Lee, Jongho Shin

AI总结:

WhisperInject通过在无害音频中嵌入细微扰动,利用两阶段对抗性攻击框架劫持音频-语言模型,生成有害内容,展示了音频原生威胁的可行性。

摘要:

随着大型语言模型(LLMs)越来越深入地融入日常生活,音频已成为人机交互的关键接口。然而,这种便利性也引入了新的漏洞,使音频成为对手攻击的潜在攻击面。我们的研究引入了WhisperInject,一种两阶段对抗性音频攻击框架,该框架操纵最先进的音频语言模型生成有害内容。我们的方法将有害载荷作为细微扰动嵌入到仍能被人类听众理解的音频输入中。第一阶段使用一种新的基于奖励的白盒优化方法,即强化学习与投影梯度下降(RL-PGD),以劫持目标模型并引发有害的原生响应。这种原生有害响应然后成为第二阶段的靶子,即载荷注入,我们使用基于梯度的优化方法将细微扰动嵌入到无害的音频载体中,如天气查询或问候信息中。我们的方法在两个基准和五个多模态LLM上实现了平均攻击成功率60-78%,并通过多个评估框架验证。我们的工作展示了一种新的实际、音频原生的威胁类别,超越了理论性的漏洞,揭示了可行且隐蔽的方法来操控多模态AI系统。

英文摘要:

As large language models (LLMs) become increasingly integrated into daily life, audio has emerged as a key interface for human-AI interaction. However, this convenience also introduces new vulnerabilities, making audio a potential attack surface for adversaries. Our research introduces WhisperInject, a two-stage adversarial audio attack framework that manipulates state-of-the-art audio language models to generate harmful content. Our method embeds harmful payloads as subtle perturbations into audio inputs that remain intelligible to human listeners. The first stage uses a novel reward-based white-box optimization method, Reinforcement Learning with Projected Gradient Descent (RL-PGD), to jailbreak the target model and elicit harmful native responses. This native harmful response then serves as the target for Stage 2, Payload Injection, where we use gradient-based optimization to embed subtle perturbations into benign audio carriers, such as weather queries or greeting messages. Our method achieves average attack success rates of 60-78% across two benchmarks and five multimodal LLMs, validated by multiple evaluation frameworks. Our work demonstrates a new class of practical, audio-native threats, moving beyond theoretical exploits to reveal a feasible and covert method for manipulating multimodal AI systems.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑