OPEN LINK ↗
// 1h agoPRODUCT UPDATE
CIRCL Refreshes VLAI Vulnerability Classifier
CIRCL’s open-source VLAI Severity model uses fine-tuned RoBERTa to classify CVE descriptions as Low, Medium, High, or Critical. Its refreshed dataset adds leakage-free evaluation splits and more consistent CVSS 4.0 labels.
// ANALYSIS
This is a practical triage accelerator, but its predictions should guide prioritization—not replace analyst-reviewed CVSS scoring.
- –Trained on more than 600,000 real-world vulnerabilities
- –Reports 74.15% accuracy and 0.6460 macro F1 on leakage-free evaluation data
- –Low-severity recall is only 32.8%, highlighting class-imbalance limitations
- –Runs locally with standard Transformers tooling and is integrated into Vulnerability-Lookup
- –Open weights and data make the pipeline reproducible for security teams
// TAGS
vlai-severitysecurityfine-tuningopen-sourceresearch
DISCOVERED
1h ago
2026-10-30
PUBLISHED
1h ago
2026-10-30
RELEVANCE
8/ 10
AUTHOR
0x0SojalSec
