TsinghuaNLP/EVIL
收藏Hugging Face2025-11-05 更新2025-11-15 收录
下载链接：
https://hf-mirror.com/datasets/TsinghuaNLP/EVIL
下载链接
链接失效反馈官方服务：
资源简介：
EVIL Dataset是一个跨越中国和美国法律环境的基准数据集，用于评估大型语言模型在面对非法用户指令时的共谋促进行为。数据集包括来自真实法庭判决的多样化非法场景和基于成熟法律框架构建的多样化非法意图，共包含5747个样本，分为中文和英文两个版本。
The EVIL (EValuation using ILlicit instructions) Dataset is a benchmark across Chinese and US legal contexts for assessing large language models complicit facilitation behaviors—cases where models enable or support illicit user instructions. It includes diverse illicit scenarios derived from real-world court judgments, combined with diverse illicit intents constructed from established legal frameworks, totaling 5,747 samples in both Chinese (zh) and English (en) versions.
提供机构：
TsinghuaNLP
