trendmicro-ailab/Primus-Seed

PRIMUS: A Pioneering Collection of Open-Source Datasets for Cybersecurity LLM Training

🤗 Primus-Seed

Primus-Seed is a high-quality🚀 cybersecurity text dataset composed of data crawled from reputable sources such as MITRE, Wikipedia, and well-known cybersecurity company websites, as well as CTI manually collected by our threat experts.

Statistics

Category	Samples	Tokens	Avg.
Web Crawl / Official Dump
Cybersecurity Blogs/News	2,946	9,751,002	3,309.9
Cybersecurity Books	6,499	2,910,464	447.8
Cybersecurity Companies Websites	76,919	65,798,561	855.4
Cybersecurity Wikipedia	6,636	9,567,196	1,441.7
MITRE	3,432	2,435,118	709.5
Expert Curation
Campaigns	136	37,106	272.8
Intrusion Sets	343	60,524	176.5
Malware	7,301	1,362,681	186.6
Reports	11,317	934,954	82.6
Threat Actors	27	2,264	83.9
Tools	238	19,926	83.7
Vulnerabilities	559,054	98,006,720	175.3
Total	674,848	190,886,516	282.9

❗❗Currently, we have only released Cybersecurity Companies Websites, Cybersecurity Wikipedia, and MITRE. Other categories are under review to ensure compliance and verify whether redistribution falls under "fair use."

How Did We Collect Cybersecurity Wikipedia?

Wikipedia does not provide a predefined cybersecurity subset, so we perform a custom filtering process. Each Wikipedia article is associated with one or more category tags, which can be further expanded into subcategory tags. Starting from the root category "Computer Security", we recursively traverse its subcategories, using GPT-4o to determine whether a category is cybersecurity-related. This process yields 375 relevant categories, from which we extract corresponding Wikipedia articles.

Prompt:

[System]
You are a helpful assistant.
[User]
Help me identify and mark the categories related to "cybersecurity", "information
security", "data protection", "cryptography", "hacker activity", "cyber attack",
"cybercrime" from a list of categories I have.
For each category, provide a reason for marking it as 'Y' (Yes) or 'N' (No) in relation to the
specified topics. Finally, output the results in JSON format with the fields: category,
reason, security.
{{category-list}

License

This dataset is released under the ODC-By license.