README.md

metadata

license: mit
task_categories:
  - text-generation
  - text-classification
  - question-answering
language:
  - en
tags:
  - cybersecurity
  - vulnerability
  - cve
  - mitre-attack
  - threat-intelligence
  - security
  - nvd
  - cisa-kev
  - infosec
size_categories:
  - 1K<n<10K
pretty_name: Cyber Threat Intelligence Dataset

Cyber Threat Intelligence Dataset

A comprehensive cybersecurity dataset combining CVE vulnerability data, MITRE ATT&CK techniques, and CISA Known Exploited Vulnerabilities — structured for AI/ML training and security research.

Author: Soham Dahivalkar
License: MIT
Created: 2026

---

Dataset Description

This dataset provides structured cybersecurity intelligence data collected from three authoritative public sources:

1. NVD (National Vulnerability Database) — CVE vulnerability records with CVSS scoring
2. MITRE ATT&CK — Enterprise attack techniques, tactics, and detection methods
3. CISA KEV — Known Exploited Vulnerabilities actively used in the wild

Each CVE record is enriched with:

• CVSS v3.1 base scores and detailed metrics
• Attack type classification (mapped from CWE)
• MITRE ATT&CK tactic mapping
• Custom risk scoring (0-100)
• CISA KEV status (actively exploited or not)
• Ransomware usage indicators

Additionally, the dataset includes instruction-tuning data for fine-tuning LLMs as cybersecurity analysts.

---

Dataset Structure

Configurations

Split	Description	Rows
cve_data	CVE vulnerability records with CVSS, CWE, risk scores	~5000
mitre_attack	MITRE ATT&CK enterprise techniques	~700
train	Instruction-tuning training split	~15000
test	Instruction-tuning evaluation split	~1500

CVE Data Schema

Column	Type	Description
cve_id	string	CVE identifier (e.g., CVE-2024-3400)
description	string	Vulnerability description
cvss_score	float	CVSS v3.1 base score (0-10)
cvss_severity	string	CRITICAL / HIGH / MEDIUM / LOW
attack_vector	string	NETWORK / ADJACENT / LOCAL / PHYSICAL
attack_complexity	string	LOW / HIGH
attack_type	string	Mapped from CWE (e.g., SQL Injection, Buffer Overflow)
cwe_ids	string	CWE weakness identifiers
risk_score	float	Custom composite risk score (0-100)
risk_level	string	CRITICAL / HIGH / MEDIUM / LOW / INFO
exploit_available	bool	Whether public exploits exist
in_cisa_kev	bool	Whether listed in CISA KEV catalog
ransomware_use	string	Known ransomware campaign usage
likely_mitre_tactic	string	Mapped MITRE ATT&CK tactic
affected_products	string	Vendor/product affected

Instruction Data Schema

Column	Type	Description
instruction	string	The task instruction
input	string	CVE or technique context
output	string	Detailed expert analysis

Instruction types include:

• CVE vulnerability analysis
• Remediation recommendations
• Risk scoring assessments
• MITRE ATT&CK mapping
• Triage prioritization decisions
• MITRE technique explanations

---

Usage

Load the Dataset

from datasets import load_dataset

# Load all splits
dataset = load_dataset("soham-dahivalkar/cyber-threat-intelligence")

# Access CVE data
cve_data = dataset["cve_data"]
print(f"Total CVEs: {len(cve_data)}")
print(cve_data[0])

# Access MITRE techniques
mitre = dataset["mitre_attack"]
print(f"Total techniques: {len(mitre)}")

# Access training data
train = dataset["train"]
print(f"Training samples: {len(train)}")
print(train[0]["instruction"])

Filter Critical Vulnerabilities

critical_cves = cve_data.filter(lambda x: x["risk_level"] == "CRITICAL")
print(f"Critical CVEs: {len(critical_cves)}")

Get Actively Exploited CVEs

exploited = cve_data.filter(lambda x: x["in_cisa_kev"] == "True")
print(f"Actively exploited CVEs: {len(exploited)}")

Use for Fine-Tuning

# Ready-to-use instruction format
for sample in dataset["train"]:
    instruction = sample["instruction"]
    input_text = sample["input"]
    output = sample["output"]
    # Format for your model and train!

---

Data Sources

Source	URL	License
NVD (National Vulnerability Database)	https://nvd.nist.gov	Public Domain
MITRE ATT&CK	https://attack.mitre.org	Apache 2.0
CISA KEV Catalog	https://www.cisa.gov/known-exploited-vulnerabilities-catalog	Public Domain

All data is collected from publicly available, free government and community sources.

---

Intended Uses

• Fine-tuning LLMs for cybersecurity analysis tasks
• Training classifiers for vulnerability severity prediction
• Building RAG systems for security knowledge retrieval
• Research on automated vulnerability assessment
• Education on cybersecurity threat intelligence

Limitations

• CVE descriptions are sourced from NVD and may not reflect the latest updates
• Risk scores are computed using a custom formula and may differ from organizational assessments
• MITRE ATT&CK mappings from CWE are approximate and based on common associations
• The instruction-tuning data is synthetically generated from structured fields

---

About the Author

Soham Dahivalkar — Generative AI Engineer specializing in agentic AI systems, enterprise RAG, and cybersecurity intelligence.

• Published Author: "Generative AI: High Stakes Cyber Security" (Amazon Kindle)
• Research: "AI in Security: ML Approach for Vulnerability Management" (ResearchGate)
• Open Source: ai-bridge-kit — Unified Python SDK for AI Providers (PyPI)
• Experience: Alembic Pharmaceuticals, CyberNX Technologies, TalaKunchi Networks
• LinkedIn: Soham Dahivalkar

---

Citation

@dataset{dahivalkar2026cyberthreat,
  author = {Dahivalkar, Soham},
  title = {Cyber Threat Intelligence Dataset},
  year = {2026},
  publisher = {HuggingFace},
  url = {https://huggingface.co/datasets/soham-dahivalkar/cyber-threat-intelligence}
}