Evidence · DiggingBeagle record
Sirens’Whisper: Inaudible Near-Ultrasonic Jailbreaks of Speech-Driven LLMs
Evidence record
Claim-level citations (2)
- supportsSirens’Whisper injected near-ultrasonic prompts into speech-driven LLMs: The authors report black-box effectiveness against commercial and open-source speech-driven LLMs and a user study in which the injected audio was perceptually indistinguishable from background-only playback.
Abstract, evaluation and user-study summary
- supportsSirens’Whisper injected near-ultrasonic prompts into speech-driven LLMs: Sirens’Whisper encodes target baseband audio into near-ultrasonic waveforms that can demodulate after acoustic transmission and microphone nonlinearity, enabling covert prompts to speech-driven LLMs.
Abstract