license: cc-by-nc-sa-4.0
language:
- en
tags: - security
- cybersecurity
- dast
- vulnerability
- penetration-testing
- owasp
- web-security
- appsec
- red-team
size_categories: - 100K<n<1M
task_categories: - text-generation
pretty_name: "Brahmastra DarkNetra"
dataset_info:
features:- name: messages
dtype: string - name: source
dtype: string - name: license
dtype: string - name: category
dtype: string - name: turns
dtype: int32
- name: messages
Brahmastra DarkNetra
The Dark Eye that sees every vulnerability in the shadows.
Security Research Dataset Notice: This dataset contains cybersecurity training data
including descriptions of vulnerability exploitation techniques, security testing payloads,
and attack methodologies for educational and defensive purposes. Antivirus software may
flag files due to pattern matching on security-related text. This is expected behavior
for cybersecurity datasets and the files do NOT contain executable malware. Raw exploit
payloads have been sanitized to reduce false positives while preserving educational value.
The largest open-source ChatML dataset for training AI-powered security scanners and cybersecurity assistants. Named from Sanskrit: Netra = eye, vision. DarkNetra sees what others miss.
Dataset Summary
| Metric | Value |
|---|---|
| Total samples | 838,679 |
| Train split | 813,519 |
| Test split | 25,160 |
| Format | ChatML (messages array) |
| License | CC BY-NC-SA 4.0 |
What's Inside
| Category | Samples | Description |
|---|---|---|
| cve-knowledge | 346,212 | |
| pentest-methodology | 318,809 | |
| exploit-knowledge | 94,870 | |
| synthetic-dast | 49,500 | |
| cybersec-reasoning | 31,853 | |
| bugbounty | 10,432 | |
| scanner-intelligence | 10,261 | |
| attack-methodology | 6,590 | |
| tool-knowledge | 5,784 | |
| attack-knowledge | 4,426 | |
| owasp-standards | 3,976 | |
| lab-writeups | 1,203 |
Sources
| Source | License | Samples |
|---|---|---|
| Pentest-Agent | Apache-2.0 | 318,809 |
| All-CVE-Records | CC-BY-4.0 | 297,346 |
| BRAHMASTRA-Synthetic | MIT | 49,500 |
| NVD-Web-CVEs | CC-BY-4.0 | 48,866 |
| ExploitDB | GPL-2.0 | 48,048 |
| ExploitDB-WAF | MIT | 46,822 |
| Cybersec-32K | Apache-2.0 | 31,853 |
| BugBounty-Public | CC-BY-NC-SA-4.0 | 10,432 |
| Scanner-OSS | Mixed-OSS | 10,261 |
| HackTricks | CC-BY-NC-4.0 | 6,590 |
| Tool-Intelligence | MIT | 5,784 |
| PayloadsAllTheThings | MIT | 4,426 |
| OWASP-Top10 | CC-BY-SA-4.0 | 1,594 |
| PortSwigger-Community | CC-BY-NC-SA-4.0 | 1,203 |
| OWASP-WSTG | CC-BY-SA-4.0 | 998 |
| OWASP-MAS | CC-BY-SA-4.0 | 813 |
| OWASP-ASVS | CC-BY-SA-4.0 | 382 |
| OWASP-SAMM | CC-BY-SA-4.0 | 189 |
Sample Format
{
"messages": [
{"role": "system", "content": "You are a cybersecurity expert..."},
{"role": "user", "content": "How do you test for blind SQL injection?"},
{"role": "assistant", "content": "<think>Blind SQLi requires...</think>..."}
],
"_source": "BRAHMASTRA-Synthetic",
"_license": "MIT",
"_category": "synthetic-dast",
"_turns": 3
}
Usage
from datasets import load_dataset
ds = load_dataset("Krishnapadala55/Brahmastra-DarkNetra")
print(f"Train: {len(ds['train']):,} | Test: {len(ds['test']):,}")
License
CC BY-NC-SA 4.0 - See individual _license fields per sample for source-specific terms.
Citation
@dataset{brahmastra_darknetra_2026,
title={Brahmastra DarkNetra: CyberSec DAST Dataset},
author={Krishna Padala},
year={2026},
url={https://huggingface.co/datasets/Krishnapadala55/Brahmastra-DarkNetra}
}
Built with BRAHMASTRA - Precision. Power. It never misses.