Dataset · JailbreakBench
LLM Jailbreak Benchmark
JailbreakBench/JBB-Behaviors
A small English benchmark for evaluating LLM jailbreak success rates and defenses.
What's inside
JBB-Behaviors is an open robustness benchmark for large language model jailbreaks, presented at the NeurIPS 2024 Datasets and Benchmarks Track, featuring tabular and text data. Available in English, it contains fewer than 1,000 examples.
What you can do with it
- Evaluating LLM robustness against jailbreaks
- Comparing attacks and defenses across different models
- Testing safety filters
- Languages
- English
- Size
- up to 1K rows
- Downloads this month
- 45.2K
- Likes
- 159
Models trained on it
- codeintegrity-ai/promptguardText classification 1.4K
- Horizon-Labs/content-safety-guard-smallText classification 1.1K
- mradermacher/iconoclast-llama3.1-8b-GGUF 978
- mradermacher/iconoclast-mistral-7b-GGUF 431
- katanemo/Arch-GuardText classification 396
- Horizon-Labs/content-safety-guard-baseText classification 338
- dannyliv/agent-guard-modernbert-baseText classification 284
- shesettipavankumarswamy/ahd-jailbreak-v1Text classification 129
How to load it
from datasets import load_dataset
ds = load_dataset("JailbreakBench/JBB-Behaviors")