AI RESEARCH
On Optimizing Multimodal Jailbreaks for Spoken Language Models
arXiv CS.LG
•
ArXi:2603.19127v1 Announce Type: new 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