Tree of Attacks: Jailbreaking Black-Box LLMs Automatically

While Large Language Models (LLMs) display versatile functionality, they continue to generate harmful, biased, and toxic content, as demonstrated by the prevalence of human-designed jailbreaks .