The research report “An Efficient, Fine-Tuned LLM for the Prediction of Vulnerability Risk Scores” has been selected for publication

The research report “An efficient fine-tuned LLM for the prediction of vulnerability risk score“, created by our colleagues Roberto Lorusso, Antonio Maci, Alessandro Santorsola, Pietro Spalluto, and Stefano Valcada of the BV TECH CyberLab of Grottaglie and Rutigliano, was selected for publication in the book”Lecture Notes in Artificial Intelligence“ by Springer, part of the ”Lecture Notes in Computer Science” series. The research consolidates and expands upon the scientific findings presented at the 18th edition of the “International Conference on Agents and Artificial Intelligence” (ICAART 2026) in the research paper “JewelCVSS: A Domain-Tuned LLM for Automated Vulnerability Scoring,” authored by the same researchers, whose work has already been recognized by the scientific community with the “Best Industrial Paper Award.”

The research report validates and confirms the effectiveness of the proposed fine-tuning procedure as a key factor in the performance improvements observed experimentally, regardless of the evolution of the language model (LLM), by applying the same experimental framework to the most recent iterations of the Gemma4 base models. The research was conducted as part of the activities outlined in the “Cybersecurity Product Suite and SOC” Program Contract.