LFrancis/

hazmart

This dataset card provides detailed information about the HazMart benchmark dataset.

	Dataset Details




	


	Dataset Description

HazMart is a benchmark dataset consisting of 77 human written scenarios distributed across 11 harm categories. It evaluates Large Language Model safety and faithfulness by simulating an autonomous AI shopkeeper facing various safety risks.

Curated by: Luca Joshua Francis, Dominik Meier, Marco Bernhard… See the full description on the dataset page: https://huggingface.co/datasets/LFrancis/hazmart.