mlsanityData to 7 Oct 2026

Audio jailbreaks and speech backdoors

56 papers1 this weekHigh confidence in this grouping: the papers form one clear topicHigh confidence in this grouping: the papers form one clear topic · About

Keywords: audio, speech, attack, attacks, lalms

Key facts. Audio jailbreaks and speech backdoors: 56 papers on arXiv since 2022 (data through 7 Oct 2026). 21 of its 56 mapped papers (38%) cite “AdvWave: Stealthy Adversarial Jailbreak Attack against Large Audio-Language Models” (2024, arXiv 2412.08608). 15 of its 56 mapped papers (27%) cite “Voice Jailbreak Attacks Against GPT-4o” (2024, arXiv 2405.19103). 3 new papers 24 Sep–7 Oct.

What’s new

24 Sep – 7 Oct 2026

3 new since 24 Sep

  1. Backdooring Acoustic Foundation Models for Physically Realizable Triggers
  2. Where Does the Audio Jailbreak Live? A Controlled Frequency-Depth Audit of AdvWave-P on Qwen2-Audio
  3. AEGIS: Audio Endogenous Guarding via Internal Signals Against Large Audio-Language Model Jailbreaks

Foundations

Most cited by papers here

  1. AdvWave: Stealthy Adversarial Jailbreak Attack against Large Audio-Language Models2024cited by 21 here
  2. Voice Jailbreak Attacks Against GPT-4o2024cited by 15 here
  3. Audio Is the Achilles' Heel: Red Teaming Audio Large Multimodal Models2024cited by 15 here
  4. Jailbreak-AudioBench: In-Depth Evaluation and Analysis of Jailbreak Threats for Large Audio Language Models2025cited by 12 here
  5. Multilingual and Multi-Accent Jailbreaking of Audio LLMs2025cited by 10 here

Lineage

The papers this direction builds on, oldest first.

Each paper on the line cites the one before.

  1. 2023Jailbroken: How Does LLM Safety Training Fail?cited by 12 here
  2. 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 29 here
  3. 2023AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language Modelscited by 7 here
  4. 2023Jailbreak and Guard Aligned Language Models with Only Few In-Context Demonstrationscited by 6 here
  5. 2024How Johnny Can Persuade LLMs to Jailbreak Them: Rethinking Persuasion to Challenge AI Safety by Humanizing LLMscited by 7 here
  6. 2024HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusalcited by 15 here
  7. 2024Audio Is the Achilles' Heel: Red Teaming Audio Large Multimodal Modelscited by 15 here
  8. 2025Jailbreak-AudioBench: In-Depth Evaluation and Analysis of Jailbreak Threats for Large Audio Language Modelscited by 12 here
  9. 2025Multilingual and Multi-Accent Jailbreaking of Audio LLMscited by 10 here
  10. 2025AudioJailbreak: Jailbreak Attacks against End-to-End Large Audio-Language Modelscited by 8 here
14 more it builds on, not in this chain
  1. 2023Fake the Real: Backdoor Attack on Deep Speech Classification via Voice Conversioncited by 7 here
  2. 2023Towards Stealthy Backdoor Attacks against Speech Recognition via Elements of Soundcited by 8 here
  3. 2023"Do Anything Now": Characterizing and Evaluating In-The-Wild Jailbreak Prompts on Large Language Modelscited by 8 here
  4. 2023Jailbreaking Black Box Large Language Models in Twenty Queriescited by 13 here
  5. 2023FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Promptscited by 6 here
  6. 2023Tree of Attacks: Jailbreaking Black-Box LLMs Automaticallycited by 8 here
  7. 2024JailbreakBench: An Open Robustness Benchmark for Jailbreaking Large Language Modelscited by 7 here
  8. 2024SpeechGuard: Exploring the Adversarial Robustness of Multimodal Large Language Modelscited by 8 here
  9. 2024Voice Jailbreak Attacks Against GPT-4ocited by 15 here
  10. 2024Unveiling the Safety of GPT-4o: An Empirical Study using Jailbreak Attackscited by 6 here
  11. 2024Best-of-N Jailbreakingcited by 7 here
  12. 2024AdvWave: Stealthy Adversarial Jailbreak Attack against Large Audio-Language Modelscited by 21 here
  13. 2025Audio Jailbreak: An Open Comprehensive Benchmark for Jailbreaking Large Audio-Language Modelscited by 9 here
  14. 2025JALMBench: Benchmarking Jailbreak Vulnerabilities in Audio Language Modelscited by 8 here

The 24 most cited of 220.

Browse

30 of 56, ranked 6–35

  1. 6Audio Jailbreak: An Open Comprehensive Benchmark for Jailbreaking Large Audio-Language Models20259
  2. 7AudioJailbreak: Jailbreak Attacks against End-to-End Large Audio-Language Models20258
  3. 8JALMBench: Benchmarking Jailbreak Vulnerabilities in Audio Language Models20258
  4. 9SpeechGuard: Exploring the Adversarial Robustness of Multimodal Large Language Models20248
  5. 10Towards Stealthy Backdoor Attacks against Speech Recognition via Elements of Sound20238
  6. 11Fake the Real: Backdoor Attack on Deep Speech Classification via Voice Conversion20237
  7. 12Speech-Audio Compositional Attacks on Multimodal LLMs and Their Mitigation with SALMONN-Guard20256
  8. 13VSVC: Backdoor attack against Keyword Spotting based on Voiceprint Selection and Voice Conversion20225
  9. 14"I am bad": Interpreting Stealthy, Universal and Robust Audio Jailbreaks in Audio-Language Models20254
  10. 15PBSM: Backdoor attack against Keyword spotting based on pitch boosting and sound masking20224
All 30
  1. 16`From Prompt to Perturbation': An Adaptive Framework for Voice-Based Jailbreaks on Audio LLMs20253
  2. 17SPIRIT: Patching Speech Language Models against Jailbreak Attacks20253
  3. 18Imperceptible Rhythm Backdoor Attacks: Exploring Rhythm Transformation for Embedding Undetectable Vulnerabilities on Speech Recognition20243
  4. 19Audio Jailbreak Attacks: Exposing Vulnerabilities in SpeechGPT in a White-Box Framework20252
  5. 20Hidden in the Noise: Unveiling Backdoors in Audio LLMs Alignment through Latent Acoustic Pattern Triggers20252
  6. 21When Good Sounds Go Adversarial: Jailbreaking Audio-Language Models with Benign Inputs20252
  7. 22AudioGuard: Toward Comprehensive Audio Safety Protection Across Diverse Threat Models20261
  8. 23Hijacking Large Audio-Language Models via Context-Agnostic and Imperceptible Auditory Prompt Injection20261
  9. 24Sparse Tokens Suffice: Jailbreaking Audio Language Models via Token-Aware Gradient Optimization20261
  10. 25Acoustic Interference: A New Paradigm Weaponizing Acoustic Latent Semantic for Universal Jailbreak against Large Audio Language Models20261
  11. 26ERIS: Evolutionary Real-world Interference Scheme for Jailbreaking Audio Large Models20251
  12. 27ALMGuard: Safety Shortcuts and Where to Find Them as Guardrails for Audio-Language Models20251
  13. 28Clean Label Attacks against SLU Systems20241
  14. 29Defend Data Poisoning Attacks on Voice Authentication20221
  15. 30Sirens' Whisper: Inaudible Near-Ultrasonic Jailbreaks of Speech-Driven LLMs20260
  16. 31On Optimizing Multimodal Jailbreaks for Spoken Language Models20260
  17. 32GRM: Utility-Aware Jailbreak Attacks on Audio LLMs via Gradient-Ratio Masking20260
  18. 33Benign Fine-Tuning Breaks Safety Alignment in Audio LLMs20260
  19. 34Audio Jailbreaks in Large Audio-Language Models: Taxonomy, Attack-Defense Analysis, and Cost-Aware Evaluation20260
  20. 35FoeGlass: Simple In-Context Learning Is Enough for Red Teaming Audio Deepfake Detectors20260