Skip to content

Dataset · Necent

Jailbreaks and prompt injections for guardrails

Necent/llm-jailbreak-prompt-injection-dataset

Multilingual set for training LLM safety filters: jailbreaks and injections.

What's inside

Combined safety dataset from 30+ public sources for training LLM guardrails, moderation classifiers, and response safety filters. Each example has four orthogonal multilabel tags in the style of AI2 WildGuard, IBM Granite Guardian, and Azure Prompt Shields instead of a single binary is_dangerous. Languages by metadata: English, Chinese, Arabic, Russian, French, and German; size is 1-10 million examples.

What you can do with it

  • Train a jailbreak and prompt injection classifier
  • Build a guardrail model with multilabel markup
  • Test moderation in six languages
Tasks
Text classification
Languages
English, Chinese, Arabic, Russian, French, German, Spanish, Hindi and 4 more
Size
1M–10M rows
Downloads this month
2.9K
Likes
93
Access
on request on Hugging Face

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("Necent/llm-jailbreak-prompt-injection-dataset")