Explainable Agentic Detector Flags Conversational Scams at 97.8% Accuracy
arXiv·medium signal
An explainable agentic system (arXiv 2607.11707) detects scams at the multi-turn conversation level rather than per-message, reaching 97.8% accuracy with reasoning-grounded explanations of why a conversation is flagged. The conversation-level framing and explainability make it a template for trust-and-safety agents that must justify decisions.