ReconGuard

**ReconGuard — Cross-Conversation Social Engineering Detection for Customer Support** ReconGuard is an AI-powered security layer that helps customer support teams identify social-engineering risks involving internal organizational knowledge. Attackers can collect small pieces of information through ordinary support interactions, such as internal team names, approval responsibilities, and account recovery procedures. These disclosures may appear harmless individually, while later requests can reference the same information to claim authority. Traditional sensitive-data detection does not always capture this form of organizational reconnaissance. ReconGuard addresses this gap by using Meta Muse Spark 1.3 to extract structured, evidence-grounded observations from support transcripts. Its Python-based correlation engine validates the observations, records organizational exposure in an Exposure Ledger, and identifies later references across separate interactions without assuming that requesters share an identity or are coordinating. For sensitive requests involving unverified internal authority claims, ReconGuard produces actionable verification guidance for support agents. The hackathon prototype demonstrates four fictional interactions across chat, email, and phone transcripts. It includes actual Meta Model API analysis, evidence validation, an organization-level exposure ledger, risk assessment, and a Streamlit interface with recorded API result replay. The current prototype is a batch-based proof of concept, not a live contact-center integration. It does not automatically block transactions, verify identities, or make claims about malicious intent. ReconGuard aims to make overlooked social-engineering signals visible before they lead to harmful account actions.