Research map / Vision-language and multimodal / Cross-modal learning
Acoustic Interference: A New Paradigm Weaponizing Acoustic Latent Semantic for Universal Jailbreak against Large Audio Language Models
Yanyun Wang, Yu Huang, Zi Liang, Xixin Wu, Li Liu
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add ALS-synthesized benign interference audio to jailbreak text | AIA · JBB, Qwen2.5-Omni | ASR-M (%): 50.98 -> 100.00 | Improved |
| add ALS-synthesized benign interference audio to jailbreak text | AIA · JBB, Gemini-3-Pro | ASR-M (%): 34.48 -> 93.10 | Improved |
| add interference audio to strong jailbreak texts | AIA · JBB strong set, Qwen2.5-Omni | ASR | Hurt |
| add interference audio to weak and medium jailbreak texts | AIA · JBB weak/medium sets, Qwen2.5-Omni | ASR | Improved |
| add AIA on less harmful HH-RLHF prompts (ablation) | AIA · HH-RLHF, Gemini-2.5 | ASR-M (%): 2.94 -> 23.53 | Improved |
| add AIA on less harmful HH-RLHF prompts (ablation) | AIA · HH-RLHF, GPT-4o-Audio | ASR-M (%): 0.00 -> 1.10 | Improved |
| use top 30 ALS instead of non-ALS audio (ablation) | AIA · JBB, Qwen2.5-Omni | ASR-M (%): 85.19 -> 100.00 | Improved |
| use top 30 ALS instead of non-ALS audio (ablation) | AIA · JBB, GPT-4o-Audio | ASR-M (%): 71.11 -> 75.61 | Improved |
| use top 30 ALS instead of bottom 30 (ablation) | AIA · JBB, Qwen2.5-Omni | ASR-R (%): 75.61 -> 74.47 | No clear effect |
| use top 30 ALS instead of bottom 30 (ablation) | AIA · JBB, GPT-4o-Audio | ASR-M (%): 67.50 -> 75.61 | Improved |
| vary benign instruction content of interference audio (ablation) | AIA · JBB, Qwen2.5-Omni | ASR-M (%): 100.00 / 100.00 / 98.04 | No clear effect |
| swap Bark for Tortoise TTS as surrogate AGM (ablation) | AIA · JBB, Qwen2.5-Omni | ASR-M (%): 100.00 -> 98.04 | No clear effect |
Closest papers by meaning
- Jailbreak-AudioBench: In-Depth Evaluation and Analysis of Jailbreak Threats for Large Audio Language Models (NeurIPS 2025 · similarity 0.90)
- ALMGuard: Safety Shortcuts and Where to Find Them as Guardrails for Audio–Language Models (NeurIPS 2025 · similarity 0.90)
- AdvWave: Stealthy Adversarial Jailbreak Attack against Large Audio-Language Models (ICLR 2025 · similarity 0.86)
- Do We Really Need Curated Malicious Data for Safety Alignment in Multi-modal Large Language Models? (CVPR 2025 · similarity 0.85)
- Probe before You Talk: Towards Black-box Defense against Backdoor Unalignment for Large Language Models (ICLR 2025 · similarity 0.85)
- Adaptive Compression and Targeted Perturbation: A Unified Framework for Generalized Audio Deepfake Detection (NeurIPS 2026 · similarity 0.84)
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.
