New Classifier Maps CVEs to MITRE ATT&CK Techniques, LLM Labeling Shows No Benefit

Researchers have developed a reproducible pipeline to map Common Vulnerabilities and Exposures (CVEs) to MITRE ATT&CK Enterprise techniques using free-text descriptions. Their custom-trained classifier, built on expert-annotated data, significantly outperforms a zero-shot baseline. Investigations into using LLM-assisted labeling to expand this dataset revealed that such labels offer no reliable improvement and can even degrade performance, suggesting that label quality, not just dataset size, is the critical factor for classifier accuracy. AI

IMPACT Highlights limitations of current LLMs in generating high-quality, reliable labels for specialized cybersecurity tasks.

RANK_REASON Academic paper detailing a new methodology and experimental results. [lever_c_demoted from research: ic=1 ai=1.0]

Read on Hugging Face Daily Papers →

• ATT&CK Enterprise
• capecitabine
• Common Vulnerabilities and Exposures ID
• Common Weakness Enumeration
• MITRE ATT&CK
• MITRE Center for Threat-Informed Defense

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →