A production-grade dataset containing over 20,000,000 rows of high-fidelity synthetic malware analysis events. Built specifically for training binary malware classifiers, multi-class family attribution models, and behavioral anomaly detectors — without exposing real binaries, live telemetry, or proprietary threat intelligence.

Massive Scale: 20M+ rows delivered in Parquet format
Rich Schema: 23 columns covering static, behavioral, and contextual malware signals
Realistic Distribution: 80% malicious / 20% benign — consistent with enterprise endpoint telemetry
Multi-Vendor Validation: Detection events span 8 industry-standard scan engines
Family Coverage: 8 malware families including Ransomware, Rootkit, Trojan, Backdoor, Worm, Spyware, Dropper, and Adware

⚠️ SAFETY & FIDELITY NOTICE — Zia-Data-Labs

This dataset is high-fidelity synthetic data engineered to mirror real-world patterns with exceptional accuracy. In benchmark testing, leading AI models treat this data as authentic — recognizing behavioral signatures, flagging anomalies, and generating functional detection code with zero scrubbing required.

Because of this realism, improper use during model fine-tuning can trigger deep behavioral shifts in production systems.

This dataset is strictly intended for research, evaluation, and development within isolated sandbox or staging environments.

Zia-Data-Labs provides all datasets on an "as-is" basis. We do not assume liability for downstream model behavior, deployment risks, or production system impacts. Users are solely responsible for conducting independent safety audits prior to any live deployment.

Instant Free Sample

Test the data quality immediately. No account or signup required.

Download 50-Row Free Sample (CSV)

Try the AI Challenge

Paste this free sample into Gemini or ChatGPT. Ask the AI to run a full malware triage analysis. Both models independently identify high-confidence malicious samples, isolate behavioral outliers, and write functional Python code to visualize entropy distributions and family breakdowns. Independently scored 100/100 for realism — no scrubbing necessary.

Access & Pricing

1. Full Dataset Access — $20.00

All 20,000,000 rows on Hugging Face.

Pay securely via PayPal: Click Here to Pay
Email your Hugging Face username to: zia.data.team@protonmail.com
Your account will be whitelisted within 24 hours.

2. Custom 1 Billion Row Dataset — $499.99

Built to your exact specifications. Contact zia.data.team@protonmail.com for details.

How to Load

from datasets import load_dataset

ds = load_dataset("Zia-Data-Labs/ZiaSyntheticDataMalware")
print(ds['train'][0])

Data Schema & Fields

Field Name	Data Type	Description
`sample_id`	string	Unique record identifier (SID-XXXXXXXX)
`file_name`	string	Synthetic filename with realistic naming conventions
`file_size_bytes`	int32	File size in bytes
`file_type`	string	Detected MIME/container type
`file_extension`	string	File extension as submitted
`file_hash_md5`	string	MD5 hash (synthetic)
`file_hash_sha256`	string	SHA-256 hash (synthetic)
`label`	string	Ground truth: `malware` or `benign`
`malware_family`	string	Family classification (Trojan, Rootkit, Ransomware, etc.)
`malware_name`	string	Specific malware variant name
`confidence_score`	float32	Detection confidence score [0.0–1.0]
`registry_modifications`	boolean	Registry write activity detected
`network_activity`	boolean	Outbound or C2 network activity detected
`obfuscation_detected`	boolean	Code obfuscation or packing layer identified
`sandbox_triggered`	boolean	Behavioral sandbox alert triggered
`signature_match`	boolean	Known signature database match
`file_entropy`	float32	Shannon entropy of file contents
`is_packed`	boolean	Packer detected
`imported_api_count`	int32	Number of imported API calls
`first_seen_date`	date	Date the sample type was first observed (YYYY-MM-DD)
`execution_timestamp`	timestamp	Execution event timestamp (ISO 8601, UTC)
`source_country`	string	ISO 3166-1 alpha-2 country code of origin
`scan_engine`	string	Detection engine that processed the sample
`WM_TAG`	string	Synthetic generation watermarking tag

Technical Specifications

Format: Apache Parquet
Total Records: 20,000,000 rows
Malicious / Benign Split: 80% / 20%
Scan Engines: Bitdefender, CrowdStrike, Kaspersky, SentinelOne, Sophos, Malwarebytes, ESET, Microsoft Defender
License: Proprietary / Custom Commercial
Producer: Zia Data Labs (2026)

Strictly Prohibited: Public redistribution, resale, or mirroring of the raw Parquet files is forbidden under our commercial terms.

Citation

Zia Data Labs, 2026. https://huggingface.co/datasets/Zia-Data-Labs/ZiaSyntheticDataMalware

Downloads last month: 17

Total file size:

2.71 GB