MITRE CVE and ExploitDB: 1M+ Cybersecurity Instruction Pairs

1 to 10 million records of cybersecurity vulnerability and exploit data compiled by jason-oneal using the pentestds pipeline as of March 2026. It merges MITRE CVE, NVD CVSS scores, and ExploitDB entries into instruction-tuning formats like Alpaca and ChatML.

Use Cases

• Fine-tuning AI agents for vulnerability analysis using ChatML formatted records
• Predicting CVSS severity levels based on NVD enrichment data
• Generating pentesting Q&A pairs derived from ExploitDB entries

Strengths

• 1M to 10M record count
• Integration of MITRE, NVD, and ExploitDB sources
• Available in Alpaca, ChatML, and Harmony JSONL formats

Limitations

• Potential for mapping errors during the automated merging of ExploitDB and CVE data
• Limited to vulnerabilities documented up to the 2026 update

Provenance

Source

MITRE CVE, NVD, and ExploitDB

Collection Method

Automated pipeline (pentestds) fetching and merging external APIs and datasets

Freshness

Last updated March 2026

Geography

Global