Overthrow4232/harmful-inputs

该数据集是一个精心策划的文本样本集合，旨在识别和防止有害提示通过应用程序。它包含提示注入尝试的样本，帮助开发者在AI应用中实施基本安全措施。数据集应作为更广泛的安全策略的一部分使用，并推荐与Lakera Guard或Llama Guard等安全解决方案结合使用。数据集版本1.0包含1,045个样本，来源于多个公开数据集，并以JSONL格式发布。

This dataset is a curated collection of text samples designed to identify and prevent harmful prompts from passing through an application. It contains samples of prompt injection attempts among other data to help developers implement basic security measures in their AI applications. It is recommended to use this dataset in conjunction with other security solutions and to implement semantic search or fuzzy-matching algorithms for optimal detection results. Version 1.0 includes 1,045 samples from various public datasets, focusing on prompt injection attempts.