SoundBreak: A Systematic Study of Audio-Only Adversarial Attacks on Trimodal Models

TL;DR

This paper systematically investigates audio-only adversarial attacks on multimodal models combining audio, vision, and language, revealing vulnerabilities that can cause significant failures even with low perceptual distortions.

Contribution

It introduces a comprehensive analysis of audio-only adversarial attacks on trimodal models, highlighting a previously overlooked attack surface and evaluating attack success across multiple models and benchmarks.

Findings

Audio-only perturbations can cause up to 96% attack success rate.

Attacks remain effective at low perceptual distortions (LPIPS <= 0.08).

Transferability across models is limited, but speech systems like Whisper are highly vulnerable.

Abstract

Multimodal foundation models that integrate audio, vision, and language achieve strong performance on reasoning and generation tasks, yet their robustness to adversarial manipulation remains poorly understood. We study a realistic and underexplored threat model: untargeted, audio-only adversarial attacks on trimodal audio-video-language models. We analyze six complementary attack objectives that target different stages of multimodal processing, including audio encoder representations, cross-modal attention, hidden states, and output likelihoods. Across three state-of-the-art models and multiple benchmarks, we show that audio-only perturbations can induce severe multimodal failures, achieving up to 96% attack success rate. We further show that attacks can be successful at low perceptual distortions (LPIPS <= 0.08, SI-SNR >= 0) and benefit more from extended optimization than increased data scale. Transferability across models and encoders remains limited, while speech recognition systems such as Whisper primarily respond to perturbation magnitude, achieving >97% attack success under severe distortion. These results expose a previously overlooked single-modality attack surface in multimodal systems and motivate defenses that enforce cross-modal consistency.