Skip to content

Dataset · AlicanKiraz0

CVE vulnerability multi-turn dataset (1999–2025)

AlicanKiraz0/All-CVE-Records-Training-Dataset

Around 300 thousand CVE vulnerability records formatted as multi-turn dialogues for AI agent training.

What's inside

This collection gathers approximately 300,000 records of known vulnerabilities (CVE) published from 1999 to 2025 in English. Each record is cleaned, enhanced, and converted into multi-turn chat format. The authors position it as the largest open cybersecurity dialogue corpus.

What you can do with it

  • SFT training for specialized information security LLM assistants
  • Training and evaluation of AI agents for vulnerability analysis
Tasks
Text generation
Languages
English
Size
100K–1M rows
Downloads this month
2.9K
Likes
63

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("AlicanKiraz0/All-CVE-Records-Training-Dataset")