CONTEXTUAL THREAT INTELLIGENCE AND ALERT PRIORITIZATION WITH FOUNDATION-SEC-8B

Modern Security Operations Centers (SOCs) face an unsustainable burden:enterprises process over 2.4 million daily alerts, yet 70% are false positives, wasting analyst time and obscuring genuine threats.Traditional alert prioritization systemsrelying on static rules, isolated machine learning (ML) models, or simplistic feature engineering-struggle to contextualize alerts within complex network environments or adapt to evolving tactics.To address this, we present Foundation-Sec-8B, a domainspecific large language model (LLM) integrated into a hybrid alert prioritization framework designed to enhance cybersecurity operations.By combining contextual threat intelligence-including MITRE ATT&CK mappings and synthetic asset criticality scoring-with Random Forest classification, the framework aims to improve alert prioritization accuracy while reducing analyst workload.Evaluating alerts using the CIC-IDS2017 dataset, which simulates diverse attack scenarios and lateral movement, the methodology emphasizes contextual awareness and resource efficiency.Although empirical validation of performance metrics such as F1-score improvements will be addressed in future work, the design theoretically mitigates key security operations center (SOC) challenges by integrating adaptive modeling to reduce false positives and accelerate threat detection.Prior ablation studies indicate that