AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

Speech-Audio Compositional Attacks on Multimodal LLMs and Their Defense with SALMONN-Guard

Yudong Yang, Xuezhen Zhang, Zhifeng Han, Siyin Wang, Jimin Zhuang, Zengrui Jin, Jing Shao, Guangzhi Sun, Chao Zhang

ICML 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: LLM jailbreak and injection defense

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add SALMONN-Guard audio guard model in place of unguarded Gemini 2.5 ProGemini 2.5 Pro · SACRED-Bench overallASR (%): 66.75 -> 11.32Improved
add SALMONN-Guard instead of Gemini 2.5 Pro aloneGemini 2.5 Pro · SACRED-Bench SAOASR (%): 88.56 -> 5.16Improved
add joint audio-text pairing (MSD) versus text-only attack (ablation)MSD attack · Gemini 1.5 ProASR (%): 57.96 -> 78.58Improved
add benign text query to harmful audio versus audio-only (ablation)MSD attack · Gemini 1.5 ProASR (%): 65.03 -> 78.58Improved
increase playback speed from 1.3 to 1.5 (ablation)SSO attack · Gemini 2.5 Pro (100 samples)ASR (%): 47.00 -> 61.00Improved
decrease playback speed from 1.3 to 1.1 (ablation)SSO attack · Gemini 2.5 Pro (100 samples)ASR (%): 47.00 -> 42.00Hurt
lower volume from -8 dB to -10 dB (ablation)SSO attack · Gemini 2.5 Pro (100 samples)ASR (%): 47.00 -> 51.00Improved
raise volume from -8 dB to -6 dB (ablation)SSO attack · Gemini 2.5 Pro (100 samples)ASR (%): 47.00 -> 47.00No clear effect
shorten overlap from 500 ms to 300 ms (ablation)SSO attack · Gemini 2.5 Pro (100 samples)ASR (%): 47.00 -> 41.00Hurt
lengthen overlap from 500 ms to 700 ms (ablation)SSO attack · Gemini 2.5 Pro (100 samples)ASR (%): 47.00 -> 47.00No clear effect
add Whisper cascaded transcription defenseGemini 2.5 Pro · SACRED-Bench SSOASR (%): 37.25 -> 37.50No clear effect
add safety system prompt defenseGemini 2.5 Pro · SACRED-Bench MSDASR (%): 63.93 -> 57.77Mixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy