Agentic SABRE: An Uncertainty-Aware Neuro-Symbolic Multi-Agent Framework for Adaptive Ransomware Detection

Agentic SABRE combines semantic representation with time-window behavioral telemetry using a neuro-symbolic multi-agent architecture. It uses Monte Carlo Dropout to quantify epistemic uncertainty, allowing a decision-layer orchestrator to triage detections via risk scores and uncertainty budgets. This enables automated containment for high-confidence threats while escalating borderline cases to human analysts.