VLAI: A RoBERTa-Based Model for Automated Vulnerability Severity Classification

2025/07/04 by Cédric Bonhomme, Bonhomme, Cédric, Alexandre Dulaunoy +1 · 2 voices · 1 citation
#cs.CR

paper · pdf · doi:10.48550/arxiv.2507.03607

Abstract

This paper presents VLAI, a transformer-based model that predicts software vulnerability severity levels directly from text descriptions. Built on RoBERTa, VLAI is fine-tuned on over 600,000 real-world vulnerabilities and achieves over 82% accuracy in predicting severity categories, enabling faster and more consistent triage ahead of manual CVSS scoring. The model and dataset are open-source and integrated into the Vulnerability-Lookup service.

Cited by

• Mapping CVEs to MITRE ATT&CK Techniques: A Curated Gold-Set Classifier and the Limits of LLM-Assisted Label Expansion

Discussions

• Vulnerability Report - January 2026 [lemmy, 6 points, 0 comments]
• Vulnerability Report - June 2026 [lemmy, 5 points, 0 comments]

Related

• Automated Characterization of Software Vulnerabilities
• VADER: A Human-Evaluated Benchmark for Vulnerability Assessment, Detection, Explanation, and Remediation
• VulStamp: Vulnerability Assessment using Large Language Model
• Detecting software vulnerabilities using Language Models
• Can LLMs Classify CVEs? Investigating LLMs Capabilities in Computing CVSS Vectors