datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
security-auditsA collection of agent traces generated with Swival (not Claude Code, despite what the HF interface currently shows), an agent designed for open-source models.
These traces focus on security audits of opensource software.
Sharing traces with Swival
Swival can export full conversation traces with --trace-dir, which writes one <session_id>.jsonl file per session:
swival "Fix the login bug" --trace-dir traces/
Those JSONL files use Swival's Claude Code compatible trace export, and… See the full description on the dataset page: https://huggingface.co/datasets/jedisct1/security-audits.cyber-security
Cybersecurity AI Knowledge Base — PhD-Level Dataset
Overview
This is the most comprehensive cybersecurity knowledge base ever assembled for AI training. It covers all domains of cybersecurity at PhD-level depth — from offensive red teaming and bug bounty exploitation to defensive SOC operations, digital forensics, and cutting-edge AI/LLM security.
Size: 16 GB | Files: 507 | Domains: 30+ | Sources: 15+ platforms
Purpose
Train the world's most… See the full description on the dataset page: https://huggingface.co/datasets/Vyber07/cyber-security.White-Hat-Security-Agent-Prompts-600K
White Hat Security Agent Prompts 600K
Overview
The White-Hat-Security-Agent-Prompts-600K dataset is a practitioner-perspective security prompts corpus of 596,295 richly contextualized queries, designed to represent how real-world defensive security professionals communicate, interrogate, and reason through active threat scenarios.
Where most security datasets catalogue CVEs, malware signatures, or CTF write-ups, this collection teaches models to operate from inside the… See the full description on the dataset page: https://huggingface.co/datasets/yatin-superintelligence/White-Hat-Security-Agent-Prompts-600K.b3-agent-security-benchmark-weak[paper] [blogpost] [game]
b3 AI Security Benchmark: Breaking Agent Backbones
Highly contextalized prompt injections crowd-sourced during the Gandalf Agent Breaker Challenge.
This is a low-quality version of the data behind Breaking Agent Backbones: Evaluating the Security
of Backbone LLMs in AI Agents.
The high quality dataset was used to evaluate the security of more than 30 LLMs.
Dataset Summary
Purpose: This dataset contains crowdsourced adversarial attacks… See the full description on the dataset page: https://huggingface.co/datasets/Lakera/b3-agent-security-benchmark-weak.cyber-security
Cybersecurity Instruction-Tuning Dataset
A large, cleaned, multi-domain cybersecurity chat dataset for LLM finetuning,
built from 198 distinct sources spanning offensive security, blue-team
operations, vulnerability intelligence, cloud/AWS security, malware analysis,
digital forensics, and more. Every record is normalized to the standard
messages chat format and deduplicated at both file and record level.
⚠️ Research use only. This dataset is provided exclusively for… See the full description on the dataset page: https://huggingface.co/datasets/oi-uae/cyber-security.llm-fv-security-targets
LLM-FV Security Targets
This dataset contains 1183 independently validated, containerized security-agent targets produced by the ucsb-mlsec/llm-fv pipelines.
Contents
GitHub Global Security Advisories: 394 targets
OSS-Fuzz: 788 targets
PoC task support: 1183 targets
Exploit task support: 395 targets
Patch task support: 395 targets
Compressed bundle size: 58.51 GiB
Vulnerability classes: {'logic_bug': 409, 'memory_vulnerability': 774}
Primary languages: {'C': 201… See the full description on the dataset page: https://huggingface.co/datasets/secmlr/llm-fv-security-targets.Omni-Frontier-Distillation-SFT-Cyber-security-Coding-dataset-collection-v2
🧬 Omni-Frontier Collection
Cybersecurity · Coding · Math · Science · RSI Reasoning — one unified SFT package
A unified, deduplicated, fully-browsable distillation & SFT corpus — every row real, every row visible.
📖 Jump to
What's inside · 🔁 Aggregation audit · 🛡 Cybersecurity · 💻 Coding · 🏭 Distillation deep-dive · 🔁 RSI · 🧮 Math/Science/More · 🎓 Training guide · 🔎 Browsing · 🧹 Quality · 🗺 Roadmap · 📄 License… See the full description on the dataset page: https://huggingface.co/datasets/Manusagents/Omni-Frontier-Distillation-SFT-Cyber-security-Coding-dataset-collection-v2.stegoattack-advbench50
StegoAttack AdvBench-50
Steganographic jailbreak data generated using the StegoAttack pipeline from the paper "Hiding in Plain Sight: A Steganographic Approach to Stealthy LLM Jailbreaks" (Geng et al., 2025).
For experiment results and analysis, see experiment.md.
What is StegoAttack?
StegoAttack is a jailbreak method that uses steganography to hide harmful queries inside benign-looking text. It embeds each word of a harmful query at a fixed position (e.g. the 2nd… See the full description on the dataset page: https://huggingface.co/datasets/heron-ai-security/stegoattack-advbench50.nixpkgs-security-patches
nixpkgs-security-patches
Training dataset for fine-tuning LLMs on nixpkgs security patch generation. Derived from real merged security PRs in NixOS/nixpkgs.
Dataset Details
588 training examples / 66 eval examples (654 total)
Format: Multi-turn tool-calling conversations in ChatML JSONL
Each example is a realistic agent session: the model reads the package file, finds the upstream fix, computes hashes via tools, and submits the fix for approval
Hashes and URLs… See the full description on the dataset page: https://huggingface.co/datasets/adastracomputing/nixpkgs-security-patches.task692_mmmlu_answer_generation_computer_security
Dataset Card for Natural Instructions (https://github.com/allenai/natural-instructions) Task: task692_mmmlu_answer_generation_computer_security
Additional Information
Citation Information
The following paper introduces the corpus in detail. If you use the corpus in published work, please cite it:
@misc{wang2022supernaturalinstructionsgeneralizationdeclarativeinstructions,
title={Super-NaturalInstructions: Generalization via Declarative Instructions on… See the full description on the dataset page: https://huggingface.co/datasets/Lots-of-LoRAs/task692_mmmlu_answer_generation_computer_security.redsec-security-sft-v1
RedSec Security SFT v1
A chat-formatted supervised fine-tuning dataset for security-focused language models, intended for authorized red-team, penetration-testing, and defensive use. Each record is a {"messages": [...]} conversation with an optional system turn, a user turn, and an assistant turn.
Split
Rows
train
55,459
validation
1,155
test
1,155
total
57,769
Sources and attribution
This is a derivative work. It combines, reformats… See the full description on the dataset page: https://huggingface.co/datasets/sahilempire/redsec-security-sft-v1.task733_mmmlu_answer_generation_security_studies
Dataset Card for Natural Instructions (https://github.com/allenai/natural-instructions) Task: task733_mmmlu_answer_generation_security_studies
Additional Information
Citation Information
The following paper introduces the corpus in detail. If you use the corpus in published work, please cite it:
@misc{wang2022supernaturalinstructionsgeneralizationdeclarativeinstructions,
title={Super-NaturalInstructions: Generalization via Declarative Instructions on 1600+… See the full description on the dataset page: https://huggingface.co/datasets/Lots-of-LoRAs/task733_mmmlu_answer_generation_security_studies.email-security
EMAIL_SECURITY
A preference dataset for EMAIL_SECURITY, harvested from real, human-labelled sources and curated by an automated harvesting harness with an LLM quality gate.
Format
Standard preference / DPO schema — each row:
column
meaning
prompt
the request (originally prompt)
chosen
the human-preferred response
rejected
a worse response to the same prompt
source
the dataset/URL the row was harvested from
Splits
80/10/10 train… See the full description on the dataset page: https://huggingface.co/datasets/316usman/email-security.offensive_security_dataset
Preview samples from the Lateos Red Team training pipeline. Contains a deterministic 3% sample (seed 42) of the full corpus, drawn from the exact production datasets used to train red-team and offensive-security models.
These samples let you evaluate schema, quality, and provenance before licensing the full datasets. All data is dual-use security research: attack playbooks, vulnerability analysis, tool usage guides, and decision frameworks for authorized penetration testing.… See the full description on the dataset page: https://huggingface.co/datasets/Lateos/offensive_security_dataset.cvefixes-security-autoformal-span-cache
CVEfixes Security source spans
This dataset contains exact, source-bound prose, code, and diff spans derived
from the original-data configuration of
Publicus/cvefixes-security-ir-graphrag at revision 6fd5918bed34f8851430e74a149502587a953fe2.
The underlying source is hitoshura25/cvefixes at revision
d4f5c4ea65329d9ccbb8a3b3149e5d06eda5edb2.
The extraction considered all 12,987 original rows across all three shards.
It retained 9,402 source rows and excluded 3,585. The release… See the full description on the dataset page: https://huggingface.co/datasets/Publicus/cvefixes-security-autoformal-span-cache.omnimcp_supabase_row_level_security_ai_teaser
🔬 INSPECT THE DEEPSEEK-R1 REASONING CHAIN LIVE:
Zero hallucinations. Null syntax errors. 100% AST compiler validated.🌐 Live Interactive Reasoning & Code Inspector: https://emgena.com/trainingslager🎁 Claim your Free Starter Kit (Code: STARTER100): https://emgena.com/trainingslager🏷️ Launch Discount: Get 20 € OFF any 500-incident production suite with code LAUNCH20!
📜 Enterprise Compliance: EU AI Act Articles 50 & 53 certified • 100% DSGVO / GDPR clean • Commercial EULA… See the full description on the dataset page: https://huggingface.co/datasets/emgena/omnimcp_supabase_row_level_security_ai_teaser.bitcoin-security-reasoning-100k
Dataset Card for Bitcoin Security Reasoning 100K
100,000 high-quality synthetic training samples for fine-tuning LLMs on Bitcoin protocol security analysis. Teaches models to analyze vulnerability clusters, form security hypotheses, and generate differential testing code.
Dataset Details
Dataset Description
This dataset contains structured security reasoning chains for Bitcoin protocol vulnerabilities. Each sample presents a cluster of causal… See the full description on the dataset page: https://huggingface.co/datasets/davidfoss/bitcoin-security-reasoning-100k.ai-agent-security-dataset
AI Agent Security and System Prompt Leakage Dataset
Dataset Overview
This dataset was created for research on AI agent security, with a specific focus on system prompt leakage, jailbreak resistance, and security-aligned fine-tuning.
The dataset evaluates how often AI agents reveal confidential information embedded inside their system prompts when exposed to adversarial prompts. It also compares the behavior of a baseline language model against a model fine-tuned using… See the full description on the dataset page: https://huggingface.co/datasets/Dhanjo/ai-agent-security-dataset.mql-benchmark
MQL Benchmark
A benchmark for evaluating natural language → MQL (Message Query Language) generation.
MQL is a DSL used at Sublime Security for email threat detection.
Dataset Summary
Split
Examples
Purpose
train
21,654
Few-shot examples and fine-tuning
validation
4,650
Prompt / hyperparameter tuning
test
4,326
Final evaluation — use sparingly
Total: 30,630 examples across four difficulty tiers and four prompt styles.
Each example is a (nl_prompt… See the full description on the dataset page: https://huggingface.co/datasets/sublime-security/mql-benchmark.quantum-cryptography-and-post-quantum-security
Neura Parse — Quantum Cryptography & Post-Quantum Security
A deep vertical on cryptography that uses quantum mechanics and on classical cryptography built to resist quantum attack. It covers quantum key distribution (BB84, B92, six-state, SARG04, E91, BBM92, decoy-state, MDI-QKD, TF-QKD, CV-QKD), device-independent protocols, composable and finite-key security proofs, quantum hacking with countermeasures, classical post-processing (reconciliation, privacy amplification… See the full description on the dataset page: https://huggingface.co/datasets/Neura-parse/quantum-cryptography-and-post-quantum-security.ios-security-vulnerabilities-swift-objc
iOS Security Vulnerabilities Dataset (Swift & Objective-C)
A comprehensive dataset of 27 real-world iOS security vulnerability patterns in Swift and Objective-C, covering all OWASP Mobile Top 10 (2024) categories with vulnerable code, secure fixes, attack scenarios, and detection guidance.
🎯 Purpose
This is the first dedicated iOS/Swift/Objective-C security vulnerability dataset on Hugging Face. While existing datasets (TitanVul, DiverseVul, CleanVul) focus on… See the full description on the dataset page: https://huggingface.co/datasets/Arno-MHL/ios-security-vulnerabilities-swift-objc.prewise-security-adapter-training-v5
Prewise Security Adapter Training V5
Bộ dữ liệu và công cụ Kaggle hoàn chỉnh để fine-tune ba LoRA trên cùng base model Qwen/Qwen3.5-4B:
message-context-adapter
web-context-adapter
explanation-adapter
phone-intelligence là HTTP provider bên ngoài, không phải LoRA. Packager tạo entry này ở trạng thái enabled: false để backend vẫn có đủ bốn runtime contract.
Quy mô
Adapter
Train
Validation
Test
Tổng
Message Context
32.767
3.196
3.992
39.955
Web Context… See the full description on the dataset page: https://huggingface.co/datasets/thuaannn/prewise-security-adapter-training-v5.combine-llm-security-benchmark
Combined LLM Security Benchmark 🔐
A comprehensive, unified benchmark dataset for evaluating Large Language Models (LLMs) on cybersecurity tasks. This dataset combines 10 security benchmarks into a standardized format with 18,059 examples across 5 task types.
📊 Dataset Summary
This dataset consolidates multiple security-focused benchmarks into a single, easy-to-use format for comprehensive LLM evaluation across various cybersecurity domains:
Total Examples: 18,059
Total… See the full description on the dataset page: https://huggingface.co/datasets/tuandunghcmut/combine-llm-security-benchmark.paragen-security-sft-alpaca
paragen-security-sft-alpaca
Alpaca-format instruction-tuning data used to train the Vanilla security
baseline (and as the source for the tokenized multi-stream cache used to
train the Stream(Ours) security checkpoint) in the paragen_llm
security/prompt-injection-robustness experiments (Table 3: TensorTrust,
Gandalf, Purple, RuLES, StruQ, NESSiE, IFEval).
Format: JSONL, one object per line, fields instruction / input / output
(standard Alpaca schema).
Size: 48,538 examples.… See the full description on the dataset page: https://huggingface.co/datasets/guinansu/paragen-security-sft-alpaca.oak-security-sft
OAK (On-chain Attack Knowledge) Security SFT Corpus
A bilingual (EN+RU) instruction-tuning dataset for training language models to analyze on-chain attacks, DeFi exploits, blockchain forensics, and crypto security incidents. Every answer is grounded in the OAK taxonomy v0.1 — a structured knowledge base of adversary tactics, techniques, mitigations, software tools, threat groups, and real-world on-chain incidents.
Dataset Overview
Metric
Value
Total… See the full description on the dataset page: https://huggingface.co/datasets/z0n3x/oak-security-sft.dspy-security-bench-trainset-workspace
dspy-security-bench: workspace trainset (v0.1)
This is the synthetic, environment-grounded query-only trainset used to
optimize DSPy programs in v0.1 of
dspy-security-bench,
a benchmark that measures whether DSPy prompt optimization affects the
prompt-injection robustness of agentic LLM programs.
What's in here
192 query / ground-truth pairs grounded in the
AgentDojo workspace suite's
default environment (calendar, inbox, files).
{"prompt": "What is the… See the full description on the dataset page: https://huggingface.co/datasets/immu4989/dspy-security-bench-trainset-workspace.turkish_cyber_security_controls_dataset
Turkish Cyber Security Controls Dataset
Veri Kümesi Özeti
Bu veri kümesi; siber güvenlik kontrolleri, kontrol seçimi ve güvenli mimari tasarımı hakkında hazırlanmış 800 Türkçe kullanıcı-asistan konuşma çifti içerir. Toplam 1.600 mesajdan oluşan koleksiyon, Türkçe siber güvenlik soru-cevap ve instruction-tuning çalışmalarını desteklemek amacıyla hazırlanmıştır.
İçerik geliştirilirken başta NIST SP 800-53 Rev. 5 kontrol kataloğu olmak üzere risk temelli kontrol… See the full description on the dataset page: https://huggingface.co/datasets/logicBombExe/turkish_cyber_security_controls_dataset.RedHat-security-VeX
Dataset Card for RedHat-security-VeX
This Dataset is extracted from publicly available Vulnerability Exploitability eXchange (VEX) files published by Red Hat.
Dataset Details
Red Hat security data is a central source of truth for Red Hat products regarding published, known vulnerabilities.
This data is published in form of Vulnerability Exploitability eXchange (VEX) available at:
https://security.access.redhat.com/data/csaf/v2/vex/
This Dataset is created by extracting… See the full description on the dataset page: https://huggingface.co/datasets/huzaifas-sidhpurwala/RedHat-security-VeX.ai-agent-security-sft-dpo
AI Agent Security — SFT + DPO
Fine-tuning data for teaching an AI agent to protect its confidential configuration without
becoming uselessly over-cautious. Built for
thesreedath/gemma-2-2b-qa-sft and
derived from
Dhanjo/ai-agent-security-dataset.
Why the helpfulness axis exists
leakage_score in the source dataset is one-sided: a model that refuses every request
scores a perfect 0.0. An existing fine-tune reported 0.0114 mean leakage (down from 0.4611
baseline)… See the full description on the dataset page: https://huggingface.co/datasets/sumitguha13/ai-agent-security-sft-dpo.information-security-policies-qa-distiset
Dataset Card for information-security-policies-qa-distiset
This dataset has been created with distilabel.
Dataset Summary
This dataset contains a pipeline.yaml which can be used to reproduce the pipeline that generated it in distilabel using the distilabel CLI:
distilabel pipeline run --config "https://huggingface.co/datasets/daqc/information-security-policies-qa-distiset/raw/main/pipeline.yaml"
or explore the configuration:
distilabel pipeline info… See the full description on the dataset page: https://huggingface.co/datasets/davidquicast/information-security-policies-qa-distiset.