JailbreakBench/

JBB-Behaviors

An Open Robustness Benchmark for Jailbreaking Language Models




NeurIPS 2024 Datasets and Benchmarks Track




Paper |
Leaderboard |
Benchmark code





	


	What is JailbreakBench?

Jailbreakbench is an open-source robustness benchmark for jailbreaking large language models (LLMs). The goal of this benchmark is to comprehensively track progress toward (1) generating successful jailbreaks and (2) defending against these jailbreaks. To this end, we… See the full description on the dataset page: https://huggingface.co/datasets/JailbreakBench/JBB-Behaviors.