datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Code_Vulnerability_Security_DPO
Cybernative.ai Code Vulnerability and Security Dataset
Dataset Description
The Cybernative.ai Code Vulnerability and Security Dataset is a dataset of synthetic Data Programming by Demonstration (DPO) pairs, focusing on the intricate relationship between secure and insecure code across a variety of programming languages. This dataset is meticulously crafted to serve as a pivotal resource for researchers, cybersecurity professionals, and AI developers who are keen on… See the full description on the dataset page: https://huggingface.co/datasets/CyberNative/Code_Vulnerability_Security_DPO.vulnerability-cwe-patch
Description
This dataset, CIRCL/vulnerability-cwe-patch, provides structured, real-world vulnerabilities enriched with CWE identifiers and corresponding patches from platforms like GitHub and GitLab. It is designed to support the development of tools for vulnerability classification, triage, and automated remediation. Each entry includes metadata such as CVE/GHSA ID, a description, CWE categorization, and links to verified patch commits with associated diff content and commit… See the full description on the dataset page: https://huggingface.co/datasets/CIRCL/vulnerability-cwe-patch.code-security-vulnerability-dataset
Code Security Vulnerability Dataset
A curated multi-language dataset of 175,419 code samples labeled with 31 vulnerability classes (30 CWEs + safe) for training multi-label code vulnerability detection models. Labels are mapped to OWASP Top 10 2021 categories.
Dataset Details
Property
Value
Total Samples
175,419
Train / Val / Test
140,335 / 17,542 / 17,542
Languages
C, C++, Python, JavaScript, Java, PHP, Go
Labels
31 (multi-label)
Format
Parquet with… See the full description on the dataset page: https://huggingface.co/datasets/ayshajavd/code-security-vulnerability-dataset.vulnerability-scores
vulnerability-scores
This dataset comprises 798,886 real-world vulnerabilities used to train and evaluate VLAI,
a transformer-based model designed to predict software vulnerability severity levels directly from text descriptions,
enabling faster and more consistent triage.
The dataset is presented in the paper VLAI: A RoBERTa-Based Model for Automated Vulnerability Severity Classification.
Sources
Source
Label
Entries
Share
cvelistv5
CVE Program… See the full description on the dataset page: https://huggingface.co/datasets/CIRCL/vulnerability-scores.Code_Vulnerability_Labeled_Dataset
Dataset Card for Code_Vulnerability_Labeled_Dataset
Dataset Summary
This dataset provides (code, vulnerability) pairs. The vulnerability field takes values according to the CWE annotation:
CWE
Description
CWE-020
Improper Input Validation
CWE-022
Improper Limitation of a Pathname to a Restricted Directory (“Path Traversal”)
CWE-078
Improper Neutralization of Special Elements used in an OS Command (“OS Command Injection”)
CWE-079
Improper Neutralization of… See the full description on the dataset page: https://huggingface.co/datasets/lemon42-ai/Code_Vulnerability_Labeled_Dataset.vul-mine-vulnerability-dataset
VulMine vulnerability dataset
VulMine is a strict, naturally imbalanced function/method-level vulnerability
dataset mined from public OSV advisories and immutable public Git revisions.
The default configuration covers Java, Python, and JavaScript. The c-cpp
configuration is a C/C++ language control built with the same VulMine label and
cleaning principles for comparisons with Big-Vul.
Release
The default configuration contains vulmine-clean-v1.2.
Split… See the full description on the dataset page: https://huggingface.co/datasets/ryal-xyz/vul-mine-vulnerability-dataset.vulnerability
Dataset Card for Dataset Name
This dataset has been generated with:
https://github.com/vulnerability-lookup/VulnTrain
Based on data from the Vulnerability-Lookup instance operated by CIRCL:
https://vulnerability.circl.lu/
The dataset is derived from CVE data provided by NIST and enriched with information from the CVE Program, FKIE, and Vulnrichment.
Dataset Details
Dataset Description
Curated by: [More Information Needed]
Funded by [optional]:… See the full description on the dataset page: https://huggingface.co/datasets/CIRCL/vulnerability.Smart_Contract_Vulnerability_DatasetSmart Contract Vulnerability Dataset
Overview
The Smart Contract Vulnerability Dataset (SCV-1-2000) is a comprehensive JSONL dataset containing 2000 entries (SCV-1 to SCV-2000) focused on advanced and unconventional smart contract vulnerabilities and attack vectors, with an emphasis on Decentralized Finance (DeFi) protocols. This dataset is designed for cybersecurity professionals, blockchain developers, machine learning engineers, and data scientists to train models, evaluate… See the full description on the dataset page: https://huggingface.co/datasets/darkknight25/Smart_Contract_Vulnerability_Dataset.vulnerability-attack-techniques
vulnerability-attack-techniques
This dataset maps 1,207 CVEs to MITRE ATT&CK (Enterprise) techniques, joining
hand-curated mappings from the MITRE Center for Threat-Informed Defense (CTID)
with vulnerability descriptions from
CIRCL/vulnerability-scores.
It is intended for training and evaluating models that suggest candidate ATT&CK
techniques from a vulnerability description: CVSS tells you how bad a
vulnerability is, CWE what kind of flaw it is — ATT&CK tells defenders what… See the full description on the dataset page: https://huggingface.co/datasets/CIRCL/vulnerability-attack-techniques.Code-Vulnerability-FineTune
🔐 Code Vulnerability FineTome — CWE-Enriched Conversation Dataset
📌 Overview
This dataset converts raw security-labeled C/C++ code samples into instruction-following conversation pairs suitable for fine-tuning large language models (LLMs) on software vulnerability detection and analysis.
It is built by preprocessing and transforming the ChamaraVishwajithRajapaksha/Code_Vulnerability_Dataset (330k rows, sourced from DiverseVul + MITRE CWE enrichment) into… See the full description on the dataset page: https://huggingface.co/datasets/ChamaraVishwajithRajapaksha/Code-Vulnerability-FineTune.unified-vulnerability-intelligence-dataset
Unified Vulnerability Intelligence Dataset (UVID) v3.0 — Cyber Security Knowledge Graph
UVID is a structured cyber security knowledge graph that unifies multiple
vulnerability classification frameworks into a single knowledge base. Each of the
250 records describes one application/software security vulnerability and links
it — where authoritative data exists — across CWE, CAPEC, MITRE ATT&CK, CVSS,
14 OWASP projects, secure-fix intelligence, detection surfaces, programming… See the full description on the dataset page: https://huggingface.co/datasets/ismailtasdelen/unified-vulnerability-intelligence-dataset.SecCoderX_Reasoning_Vulnerability_Detection_SFT_Cold_Start_Dataset
Citation
If you find our work helpful, feel free to give us a cite.
@misc{wu2026securecodegenerationonline,
title={Secure Code Generation via Online Reinforcement Learning with Vulnerability Reward Model},
author={Tianyi Wu and Mingzhe Du and Yue Liu and Chengran Yang and Terry Yue Zhuo and Jiaheng Zhang and See-Kiong Ng},
year={2026},
eprint={2602.07422},
archivePrefix={arXiv},
primaryClass={cs.CR},
url={https://arxiv.org/abs/2602.07422}… See the full description on the dataset page: https://huggingface.co/datasets/SecCoderX/SecCoderX_Reasoning_Vulnerability_Detection_SFT_Cold_Start_Dataset.vulnerability-intelligence-diagrammatic-reasoning
Vulnerability Intelligence with Diagrammatic Reasoning
[!Important]
This dataset was created as a proof-of-concept for the Reasoning Datasets Competition (May 2025). If you have any feedback or suggestions, please feel free to open a discussion! Access the Github repository here.
A. Overview
This dataset focuses on security vulnerability analysis through a multi-dimensional approach that combines four types of reasoning to generate valuable insights for… See the full description on the dataset page: https://huggingface.co/datasets/davidquicast/vulnerability-intelligence-diagrammatic-reasoning.Vulnerability-CNVD
Vulnerability-CNVD
Vulnerability descriptions and severity labels from the China National Vulnerability Database (CNVD), extracted via Vulnerability-Lookup.
Dataset structure
Field
Type
Description
id
string
CNVD identifier (e.g., CNVD-2025-03529)
title
string
Vulnerability title in Chinese
description
string
Vulnerability description in Chinese
severity
string
Severity level: 高 (High), 中 (Medium), or 低 (Low)
cve_id
string
Corresponding CVE… See the full description on the dataset page: https://huggingface.co/datasets/CIRCL/Vulnerability-CNVD.Python_Vulnerability_Remediation
Python SAST Vulnerability and Remediation Dataset
Summary
This dataset is a collection of Python code snippets containing common security vulnerabilities, paired with their corresponding high-quality remediations. It is designed for fine-tuning language models to assist with Static Analysis Security Testing (SAST) by suggesting secure code fixes.
The dataset is primarily focused on vulnerabilities from the following Common Weakness Enumerations (CWEs):
CWE-89 (SQL… See the full description on the dataset page: https://huggingface.co/datasets/cmonplz/Python_Vulnerability_Remediation.java_vulnerability_datasetswe-vulnerability-bench
SWE-Vulnerability-Bench
Бенчмарк для оценки поиска уязвимостей и качества исправлений на основе SWE-bench c аннотированным датасетом (gold patch, логи тестов, модельные патчи, флаги resolved).
Цифры (по gpt-5-mini_swe-bench_vulerablities_dataset_latest.pkl)
Метрика
Значение
Всего инстансов
474
Успешно решённых (resolved)
255 (~53.8%)
Не решённых
219 (~46.2%)
Часть инстансов была потеряна при скоринге (исходный SWE-bench — порядка 500); в датасете… See the full description on the dataset page: https://huggingface.co/datasets/N1ksw1r/swe-vulnerability-bench.vulnerability-severity-classificationThis dataset combines vulnerable functions (scraped from 5 auditting companies: Codehawks, ConsenSys, Cyfrin, Sherlock, Trust Security) and auddited functions with no vulnerabilities (scraped from Etherscan)
The purpose of the dataset is to enable training of classification models to discriminate between the 4 classes: none, low, medium and high.
Field
Description
1. function
Raw solidity code
2. severity
Severity of vulnerability ('none', low, medium, high)
Data… See the full description on the dataset page: https://huggingface.co/datasets/msc-smart-contract-auditing/vulnerability-severity-classification.OpenVul_Rejection_Sampling_based_Vulnerability_Reasoning_Dataset_for_SFTThis dataset provides high-quality, correctness-filtered vulnerability reasoning data to support the SFT of specialized VD LLMs for future research.
vulnerability-attack-techniques-llm-scaling
vulnerability-attack-techniques-llm-scaling
⚠️ The labels in this dataset are machine-generated by an LLM, not
analyst-curated — and the paper that produced them found they do not
improve a classifier trained on the expert gold set. It is published for
reproducibility and for research on LLM-assisted labeling. For training,
use the curated gold set
CIRCL/vulnerability-attack-techniques.
This dataset contains 984 CVEs labeled with MITRE ATT&CK (Enterprise)
techniques by… See the full description on the dataset page: https://huggingface.co/datasets/CIRCL/vulnerability-attack-techniques-llm-scaling.vulnerability-reports
SPECA Vulnerability Reports
A normalized corpus of public smart-contract audit findings, curated by
the SPECA project. Each row is
one expert-judged audit issue, deduplicated across public competitive-audit
platforms.
The dataset is split into one config per security domain (defi,
lending, oracle, …). HuggingFace auto-detects each <domain>/
folder as a config; pick the domain you want at load time.
Loading
from datasets import load_dataset
# A specific domain… See the full description on the dataset page: https://huggingface.co/datasets/NyxFoundation/vulnerability-reports.arcs-authority-vulnerability
ARCS Authority Vulnerability Evaluation Dataset v1.1
Description
Empirical evaluation data measuring authority vulnerability in AI systems. Covers single-model evaluation, two-hop agent chain propagation, and three-hop agent chain propagation across six independent AI lineages.
This is the first published dataset measuring:
Whether AI models accept false authority claims under adversarial pressure
Whether authority vulnerability propagates between models in… See the full description on the dataset page: https://huggingface.co/datasets/aa8899/arcs-authority-vulnerability.SARD-Vulnerability-Datasetvulnerability-scores-cvss-v3
Vulnerability scores (CVSS v3 combined)
A labeled slice of CIRCL/vulnerability-scores for training and evaluating models that predict CVSS v3 severity from a vulnerability description.
Every row has a combined v3 score and a severity band. Rows with no v3.1 or v3.0 score were dropped.
What changed from the original
The CIRCL dataset stores four separate CVSS columns (cvss_v4_0, cvss_v3_1, cvss_v3_0, cvss_v2_0). Those versions are not on the same scale, so this… See the full description on the dataset page: https://huggingface.co/datasets/AgileRLArena/vulnerability-scores-cvss-v3.Code-Vulnerability-Balanced
Code Vulnerability Balanced — CWE-Enriched Conversation Dataset
📌 Overview
This dataset is a balanced and shuffled version of
ChamaraVishwajithRajapaksha/Code-Vulnerability-FineTune,
which itself was derived from the original
ChamaraVishwajithRajapaksha/Code_Vulnerability_Dataset
(330k rows, sourced from DiverseVul + MITRE CWE enrichment).
The original fine-tuning dataset was imbalanced — the number of Vulnerable and Safe
samples were not equal — and the samples were not… See the full description on the dataset page: https://huggingface.co/datasets/ChamaraVishwajithRajapaksha/Code-Vulnerability-Balanced.Code_Vulnerability_Dataset
🔐 Code Vulnerability Dataset (CWE-Enriched)
📌 Overview
This dataset is built from the bstee615/diversevul dataset and enhanced with structured vulnerability intelligence from the MITRE Common Weakness Enumeration (CWE) database.
It provides a rich, machine-readable representation of software vulnerabilities, mapping raw vulnerable code samples to standardized CWE classifications.
The dataset is designed for research and development in:
Vulnerability detection models… See the full description on the dataset page: https://huggingface.co/datasets/ChamaraVishwajithRajapaksha/Code_Vulnerability_Dataset.smart-contract-vulnerability-benchmarks
Smart Contract Vulnerability Benchmarks
smartbugs-wild/contracts is stored as lossless CSV shards with content_base64; other benchmark files are raw.
Converted smartbugs-wild contracts: 47398
Raw files: 5295
Generated: 2026-06-24T16:06:36.674733+00:00
VAB-vulnerability-analysis-benchmark
FBE and VAB
Two small benchmarks for security code analysis. Both grade without an LLM judge, so runs are cheap
and repeatable.
FBE (find-the-bug)
14 code snippets, each with one planted vulnerability. Ask the model to analyze the code, then check
whether it actually found the flaw.
Grading uses concept groups: the answer has to contain at least one synonym from every required group.
Four numbers come out:
found, did it identify the real vulnerability (this is… See the full description on the dataset page: https://huggingface.co/datasets/MK4-Research/VAB-vulnerability-analysis-benchmark.circl-vulnerability-attack-techniques
Mirror: CIRCL/vulnerability-attack-techniques
Pinned snapshot / mirror of CIRCL/vulnerability-attack-techniques, re-hosted for PROTISEC
research reproducibility. Redistributed under the upstream license (cc-by-4.0)
with attribution — all credit to the original author.
Original author: CIRCL
Source dataset: CIRCL/vulnerability-attack-techniques
License: cc-by-4.0
Family: cve
Mode: full
Rows cached: 1208
Changes vs upstream: cached snapshot, possibly stream-capped and/or… See the full description on the dataset page: https://huggingface.co/datasets/ansulev/circl-vulnerability-attack-techniques.java-vulnerability