This research introduces SE-OmniGuard, a comprehensive framework for detecting sophisticated social engineering attacks in multi-turn conversations powered by LLMs.

• Creates realistic simulations of personalized social engineering attacks
• Develops detection mechanisms that achieve 87.5% accuracy in identifying malicious conversations
• Reveals how attackers exploit personal information to build trust before executing attacks
• Demonstrates the evolving nature of threats as LLMs become more capable conversational agents

This work addresses critical security vulnerabilities in an era where AI-powered chatbots could automate and scale sophisticated social engineering attacks, providing organizations with practical detection tools and defense strategies.

Personalized Attacks of Social Engineering in Multi-turn Conversations -- LLM Agents for Simulation and Detection