leeroy-jankins/NIST-Managing-AI-Misuse-Risk
收藏资源简介:
该数据集包含从美国国家标准与技术研究院(NIST)于2025年1月发布的第二份公开草案《NIST AI 800-1 2pd:双用途基础模型的滥用风险管理》中衍生的问答记录。源文档为双用途基础模型的滥用风险管理提供自愿性指导,重点关注这些模型可能被故意滥用以危害公共安全或国家安全的威胁,包括涉及化学、生物、放射或核威胁、网络攻击、有害合成内容、儿童性虐待材料和非自愿亲密图像等滥用行为。数据集设计用于文档基础的问答、监督微调、检索增强生成评估、AI安全培训和AI治理分析。每个记录包含自然语言的用户问题和基于源指导的助理回答,旨在支持开发和评估能够利用权威联邦指导回答关于双用途基础模型滥用风险管理系统。
This dataset contains question-and-answer records derived from NIST AI 800-1 2pd, Managing Misuse Risk for Dual-Use Foundation Models, a second public draft issued by the U.S. AI Safety Institute at the National Institute of Standards and Technology in January 2025. The source document provides voluntary guidance for improving the safety, security, and trustworthiness of dual-use foundation models, focusing on the risk that such models may be deliberately misused to cause harm to public safety or national security, including misuse involving chemical, biological, radiological, or nuclear threats, offensive cyber operations, harmful synthetic content, child sexual abuse material, and non-consensual intimate imagery. The dataset is designed for document-grounded question answering, supervised fine-tuning, retrieval-augmented generation evaluation, AI safety training, and AI governance analysis. Each record contains a natural-language user question and a corresponding assistant answer grounded in the source guidance.



