Necent/

llm-jailbreak-prompt-injection-dataset

A unified safety dataset combining 30+ public sources for training LLM guardrails, content moderation classifiers, and response-safety filters.

Instead of a single binary is_dangerous, every example carries four orthogonal labels matching the structure used by AI2 WildGuard, IBM Granite Guardian, and Azure Prompt Shields:

Column Type Description

prompt str The user/attack… See the full description on the dataset page: https://huggingface.co/datasets/Necent/llm-jailbreak-prompt-injection-dataset.