Tastle: Distract Large Language Models for Automatic Jailbreak Attack

Tastle proposes a distraction-based jailbreak framework to automate red teaming of large language models, achieving superior effectiveness, scalability, and transferability.