How AI describes JailbreakBench

JailbreakBench provides a centralized benchmark to study jailbreak attacks and defenses against large language models. It maintains a repository of jailbreak artifacts for reproducibility, a standardized evaluation library with a defined threat model, prompts, and scoring, and leaderboards tracking attacker success and defense performance across open-source and closed-source models. Its JBB-Behaviors dataset includes 100 misuse behaviors (plus 100 benign behaviors) to evaluate model safety, and the project aims to accelerate safety training and adapt to advances in the field.