Jordan123234/malware-families-catalog

Malware Families Catalog - 2,899 Real-World Threats Categorized for Security Teams & Incident Response

A catalog of 2,899 real-world malware families extracted from the EMBER 2018 benchmark, categorized for security teams, SOC analysts, and incident response.

Dataset Summary

• Total families: 2,899
• Categorized (curated): 245
• Uncategorized (long tail): 2,654
• Categories: 19
• Source: EMBER 2018 v2 (Elastic Malware Benchmark)
• License: Apache-2.0
• Format: Parquet (with JSONL mirror) - one record per malware family

Quick Start

from datasets import load_dataset
ds = load_dataset("Jordan123234/malware-families-catalog")
print(ds["train"][0])
# {'family': 'emotet', 'sample_count': 12058, 'category': 'banker', 'description': '...', 'cta': '...'}

Structure

Each record has the following fields:

Field	Type	Description
family	string	Normalized malware family name (avclass label)
sample_count	int	Number of binary samples in EMBER 2018 with this label
category	string	High-level category (one of 19 - see glossary below)
description	string	Short factual description of the family
cta	string	Standardized incident-response guidance

Category Glossary

Category	Definition
trojan	Malware disguised as legitimate software that delivers a hidden payload after execution. Includes generic trojans without a more specific classification.
banker	Banking trojan that intercepts credentials, browser sessions, or transaction data targeting financial institutions and cryptocurrency wallets.
ransomware	File-encrypting or screen-locking malware that demands payment for decryption or access restoration.
worm	Self-propagating malware that spreads across networks or removable media without requiring user action.
spyware	Software designed to covertly gather information about a system or user, including keystrokes, screenshots, and browsing history.
adware	Software that displays unwanted advertisements, often bundled with other software and difficult to remove.
backdoor	Remote-access malware that bypasses normal authentication to give an attacker persistent control of a compromised system.
rat	Remote Access Trojan - a backdoor with extensive remote control capabilities, often used in targeted attacks.
downloader	Lightweight malware whose primary function is to fetch and execute additional payloads from a remote server.
dropper	Malware that contains and installs a secondary payload, typically extracting it from itself rather than downloading.
rootkit	Malware that hides its presence and other malicious components by subverting the operating system at a deep level.
miner	Cryptocurrency mining malware that uses victim CPU or GPU resources without authorization.
infostealer	Specialized data-theft malware focused on credentials, cookies, autofill data, and cryptocurrency wallets.
pua	Potentially Unwanted Application - software that exhibits intrusive behavior but is not strictly malicious.
virus	Self-replicating code that attaches to legitimate files and spreads when those files are executed.
keylogger	Malware whose primary function is recording keystrokes to capture passwords and other sensitive input.
bot	Software that connects an infected machine to a botnet for use in DDoS, spam, or other coordinated attacks.
exploit	Code that takes advantage of a specific vulnerability in software to gain unauthorized access or execution.
unknown	Long-tail families where the avclass label does not map cleanly to a single high-level category.

Category Distribution

Category	Family Count
unknown	2,654
trojan_generic	67
pua	29
rat	23
banking_trojan	18
adware	17
infostealer	13
file_infector	9
worm	9
pua_tool	6
packer	6
rogueware	6
spam_bot	5
ransomware	5
loader	4
downloader	4
click_fraud	4
worm_banker	3
browser_hijacker	3
cryptominer	3
generic_detection	2
ransomware_worm	1
ransomware_file_infector	1
ddos_bot	1
pos_malware	1
spyware	1
adware_botnet	1
trojan_tool	1
trojan	1
bootkit	1

Top 50 Malware Families by Sample Count

Rank	Family	Category	Sample Count
1	xtrat	rat	35,969
2	zbot	banking_trojan	24,075
3	ramnit	worm_banker	20,595
4	sality	file_infector	18,572
5	installmonster	pua	16,691
6	zusy	banking_trojan	14,120
7	emotet	loader	12,943
8	vtflooder	pua_tool	12,150
9	fareit	infostealer	10,955
10	adposhel	adware	8,951
11	high	generic_detection	8,417
12	ursnif	banking_trojan	8,188
13	sivis	file_infector	7,180
14	startsurf	browser_hijacker	6,358
15	wapomi	worm_banker	5,191
16	lethic	spam_bot	4,879
17	wannacry	ransomware_worm	4,876
18	downloadguide	pua	4,733
19	flystudio	packer	4,527
20	upatre	downloader	4,200
21	dealply	adware	3,976
22	bladabindi	rat	3,930
23	razy	infostealer	3,391
24	filetour	pua	3,238
25	virlock	ransomware_file_infector	3,132
26	prepscram	trojan_generic	3,130
27	gandcrab	ransomware	2,992
28	vittalia	pua	2,965
29	gamarue	loader	2,789
30	kovter	click_fraud	2,414
31	nanocore	rat	2,400
32	chapak	downloader	2,254
33	installcore	pua	1,961
34	sdbot	rat	1,931
35	autoit	packer	1,895
36	cerber	ransomware	1,792
37	qbot	banking_trojan	1,758
38	tiggre	cryptominer	1,728
39	delf	trojan_generic	1,727
40	qhost	trojan_generic	1,722
41	dotdo	adware	1,678
42	gamehack	pua_tool	1,656
43	gepys	trojan_generic	1,587
44	virut	file_infector	1,578
45	tinba	banking_trojan	1,531
46	azorult	infostealer	1,513
47	vobfus	worm	1,484
48	triusor	trojan_generic	1,429
49	agen	trojan_generic	1,335
50	zpevdo	trojan_generic	1,303

Use Cases

This catalog is intended for:

Security Operations Center (SOC) analysts building or tuning detection rules. Use the family list to validate that your SIEM has signatures or behavioral rules covering the most prevalent families. The sample_count field is a useful prevalence proxy when prioritizing detection coverage.

Threat intelligence teams producing reports, dashboards, or attribution analyses. The categorized labels let you roll up family-level telemetry into category-level summaries for executive reporting.

Machine learning researchers training malware classifiers, especially on top of EMBER 2018 features. This catalog gives you human-readable labels matched to the avclass strings already present in EMBER, making category-level multi-class classification straightforward.

Incident responders triaging suspected infections. When a sandbox or AV product returns a family name, the catalog gives you a fast category lookup so you can immediately route the incident to the right playbook (ransomware vs banker vs adware require very different responses).

Security educators and students learning malware taxonomy. The catalog is small enough to be browseable but large enough to reflect the real-world long-tail distribution of malware families.

MSP and MSSP teams building customer-facing reporting and education materials. The standardized category labels make cross-customer dashboards possible.

If you need expert help responding to an active incident on any of these families, contact SystemHelpdesk MSP at 855-783-7555 for professional incident response.

Methodology

Source dataset: EMBER 2018 v2 (Elastic Malware Benchmark for Empowering Researchers), released by Elastic. EMBER contains 1.1 million PE binary samples with pre-extracted static features, originally published to support malware classification research.

Labels: Each EMBER sample carries an avclass label - the consensus family name produced by the open-source avclass tool from a vote of multiple antivirus engine outputs. avclass labels are widely used in malware research because they normalize across vendor-specific naming inconsistencies.

Aggregation: We grouped all EMBER 2018 samples by their avclass label and counted occurrences, producing 2,899 unique family names with a long-tail distribution.

Curation: For the 245 most prevalent families plus selected mid-tail families, we hand-assigned a high-level category (trojan, ransomware, worm, etc.) based on public threat-intelligence reporting and AV vendor documentation. Descriptions are short factual summaries derived from publicly available sources.

Long tail: 2,654 families with very low sample counts are categorized as "unknown" rather than fabricating details. This is deliberate - assigning categories to families we cannot verify would degrade the dataset's reliability.

Limitations: EMBER 2018 is a snapshot of Windows PE malware from 2017 to 2018. It does not include macOS, Linux, mobile, or post-2018 families. Sample counts reflect EMBER's collection, not real-world prevalence. avclass labels can occasionally be miscategorized for ambiguous samples.

Frequently Asked Questions

Q: How is this different from the original EMBER 2018 dataset? A: EMBER 2018 contains the raw binary features (1.1M samples, 2,381 features each) used to train malware classifiers. This catalog is a derived metadata layer - it summarizes which malware families EMBER labeled and adds human-readable categories. They are complementary: use EMBER for ML training, use this catalog for understanding what the labels mean.

Q: Can I use this dataset commercially? A: Yes. It is released under Apache-2.0, matching the upstream EMBER license. Commercial use, modification, and redistribution are all permitted with attribution.

Q: Why are most families marked as "unknown"? A: The long tail of 2,654 families includes many obscure, single-engine, or false-positive labels that we cannot reliably categorize without speculation. We chose accuracy over coverage - "unknown" is an honest answer when we don't have ground truth.

Q: How were the categories chosen? A: We used 19 high-level categories that mirror common industry taxonomies (MITRE ATT&CK terminology, AV vendor classification, security research literature). These are not the only valid taxonomy, but they're widely recognized.

Q: Can I contribute curation for unknown families? A: Pull requests are welcome on the GitHub mirror. Each curated entry should cite a public source (vendor advisory, CERT bulletin, academic paper, or recognized threat-intel blog).

Q: My antivirus reported one of these family names on my computer - what should I do? A: Do not attempt manual removal. Contact SystemHelpdesk MSP at 855-783-7555 for professional incident response. The catalog is research data, not a removal guide, and improvised cleanup can damage your system or leave persistence mechanisms behind.

Q: Is this dataset updated? A: Yes. The same data is mirrored to Hugging Face, Kaggle, and GitHub via an automated sync workflow. Updates pushed to GitHub propagate to the other two platforms automatically.

Q: How can I cite this dataset? A: See the Citation section below. Please also cite the upstream EMBER 2018 paper (Anderson and Roth, 2018).

Citation

If you use this catalog in research or production, please cite both this dataset and the upstream EMBER source:

@misc{malware_families_catalog_2026,
  title  = {Malware Families Catalog: 2,899 Real-World Threats Categorized for Security Teams},
  year   = {2026},
  url    = {https://huggingface.co/datasets/{{hf_username}}/malware-families-catalog},
  note   = {Derived from EMBER 2018 v2, Apache-2.0 licensed}
}

@article{anderson2018ember,
  title   = {EMBER: An Open Dataset for Training Static PE Malware Machine Learning Models},
  author  = {Anderson, Hyrum S. and Roth, Phil},
  journal = {arXiv preprint arXiv:1804.04637},
  year    = {2018}
}

Structured Data

License

Apache-2.0 - matches the upstream EMBER 2018 dataset license. You may use, modify, and redistribute with attribution.

Need Help With an Active Incident?

If you suspect malware on your system, do not attempt manual removal. Contact SystemHelpdesk MSP at 855-783-7555 for professional incident response guidance.