Saved in:
Bibliographic Details
Main Authors: Krishnan, Aravind, Stańczak, Karolina, Klakow, Dietrich
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2603.19127
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918398439456768
author Krishnan, Aravind
Stańczak, Karolina
Klakow, Dietrich
author_facet Krishnan, Aravind
Stańczak, Karolina
Klakow, Dietrich
contents As Spoken Language Models (SLMs) integrate speech and text modalities, they inherit the safety vulnerabilities of their LLM backbone and an expanded attack surface. SLMs have been previously shown to be susceptible to jailbreaking, where adversarial prompts induce harmful responses. Yet existing attacks largely remain unimodal, optimizing either text or audio in isolation. We explore gradient-based multimodal jailbreaks by introducing JAMA (Joint Audio-text Multimodal Attack), a joint multimodal optimization framework combining Greedy Coordinate Gradient (GCG) for text and Projected Gradient Descent (PGD) for audio, to simultaneously perturb both modalities. Evaluations across four state-of-the-art SLMs and four audio types demonstrate that JAMA surpasses unimodal jailbreak rate by 1.5x to 10x. We analyze the operational dynamics of this joint attack and show that a sequential approximation method makes it 4x to 6x faster. Our findings suggest that unimodal safety is insufficient for robust SLMs. The code and data are available at https://repos.lsv.uni-saarland.de/akrishnan/multimodal-jailbreak-slm
format Preprint
id arxiv_https___arxiv_org_abs_2603_19127
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle On Optimizing Multimodal Jailbreaks for Spoken Language Models
Krishnan, Aravind
Stańczak, Karolina
Klakow, Dietrich
Machine Learning
As Spoken Language Models (SLMs) integrate speech and text modalities, they inherit the safety vulnerabilities of their LLM backbone and an expanded attack surface. SLMs have been previously shown to be susceptible to jailbreaking, where adversarial prompts induce harmful responses. Yet existing attacks largely remain unimodal, optimizing either text or audio in isolation. We explore gradient-based multimodal jailbreaks by introducing JAMA (Joint Audio-text Multimodal Attack), a joint multimodal optimization framework combining Greedy Coordinate Gradient (GCG) for text and Projected Gradient Descent (PGD) for audio, to simultaneously perturb both modalities. Evaluations across four state-of-the-art SLMs and four audio types demonstrate that JAMA surpasses unimodal jailbreak rate by 1.5x to 10x. We analyze the operational dynamics of this joint attack and show that a sequential approximation method makes it 4x to 6x faster. Our findings suggest that unimodal safety is insufficient for robust SLMs. The code and data are available at https://repos.lsv.uni-saarland.de/akrishnan/multimodal-jailbreak-slm
title On Optimizing Multimodal Jailbreaks for Spoken Language Models
topic Machine Learning
url https://arxiv.org/abs/2603.19127