leo-bjpark/

incomplete-prompt-jailbreak

A benchmark dataset for evaluating Incomplete Prompt Jailbreak (IPJ) behavior in LLMs. This dataset provides harmful-question seeds, paraphrase metadata, and fixed trigger templates for reproducible completion-style and chat-template-style safety evaluation.

	Overview




	

Item Value

Base harmful questions 210

Harm categories 7

Trigger types used 9

simple rows 210

completion_benchmark rows 1,890 (210 x 9)

chat_template_benchmark rows 1,890… See the full description on the dataset page: https://huggingface.co/datasets/leo-bjpark/incomplete-prompt-jailbreak.