CyberLLMInstruct
收藏资源简介:
CyberLLMInstruct是一个由肯特大学研究人员创建的包含54928条指令-响应对的数据集,专注于网络安全任务,如恶意软件分析、钓鱼模拟和零日漏洞等。数据集通过多阶段过程构建,包括从多个资源中筛选和结构化数据,并与现实世界场景对齐以增强其适用性。该数据集旨在帮助研究人员系统地识别LLM中的漏洞,以加强对这些模型恶意利用的防范。
CyberLLMInstruct is a dataset containing 54,928 instruction-response pairs, created by researchers at the University of Kent, focusing on cybersecurity tasks such as malware analysis, phishing simulation, and zero-day vulnerabilities. The dataset is constructed via a multi-stage process, including screening and structuring data from multiple sources, and aligning with real-world scenarios to enhance its applicability. This dataset aims to help researchers systematically identify vulnerabilities in LLMs, so as to strengthen the defense against malicious exploitation of these models.

- 1CyberLLMInstruct: A New Dataset for Analysing Safety of Fine-Tuned LLMs Using Cyber Security Data英国肯特大学 · 2025年



