Train, triage, repeat: The AI agent changing how we fight phishing
Red Canary describes a multi-agent phishing triage system achieving 94% accuracy through orchestrated subagents: email parsing/enrichment, hybrid feature extraction combining traditional rules with NLP-powered sentiment/intent analysis, a deterministic rules engine, and hybrid AI/ML classification. The rules engine pairs atomic email metadata with extracted features to detect TTP patterns and emerging campaigns, while a reasoning subagent generates final classifications with full explainability. Critically, the ML model trains only on true/false feature values—never raw email content—then adds feature importance weights to guide AI decision-making, maintaining transparency. Analyst-driven feedback loops continuously refine detection logic while automation handles bulk triage, enabling human expertise to focus on nuanced phishing variants at 3.8M+ annual attack volumes.