Jailbreaking LLMs' Safeguard with Universal Magic Words for Text Embedding Models

Open in new window