Now You Hear Me

Audit and defend large audio-language models (LALMs) against narrative-style audio jailbreaks. Based on the 'Now You Hear Me' paper (EACL 2026), this skill helps build red-team evaluation pipelines, safety classifiers, and input-validation layers for speech-enabled AI systems. Use when: 'red team my audio model', 'test voice assistant safety', 'build audio input guardrails', 'detect narrative jailbreaks in speech', 'audit LALM safety alignment', 'harden speech interface against adversarial audio'.

ndpvt-web Updated 14 repo stars

File contents

ndpvt-web/arxiv-claude-skills commit 820fd0f6ea

Frequently asked questions

npx get-skill add ndpvt-web/now-you-hear-me