Detecting social engineering threats

Abstract

Embodiments detect social engineering threats by analyzing telephony and messaging communications to identify deceptive manipulation tactics. The system employs non-agentic threat evaluation using natural language processing, pattern matching, and statistical analysis to collect social engineering indicators and generate preliminary threat confidence scores. Candidate threats undergo agentic threat evaluation using generative artificial intelligence agents that analyze prompts containing telephony and messaging activity information. The system establishes baseline communication patterns from historical data, then identifies deviations and anomalies exceeding threshold values. Multi-vector threat assessments correlate suspicious activities across communication channels using timestamps to detect coordinated campaigns. The system identifies communications demonstrating personal knowledge references, false authority claims, trust-building attempts, and coordinated manipulation strategies. Validation checks cross-reference telephone numbers against databases of known fraudulent sources and disposable communication services.