The Synthesis of Deception: AI-Driven Visual Personification and Adaptive Phishing in Modern Social Engineering
Year: 2026
Type: conference-paper
Abstract: Recent advances in generative artificial intelligence have transformed phishing from static messages into adaptive interaction processes that evolve based on user responses. However, existing studies primarily evaluate phishing in isolated settings and lack controlled environments for analyzing how adaptive strategies change over repeated interactions. This study presents a persona-aware, agent-based simulation framework for the systematic evaluation of adaptive phishing behavior under controlled experimental conditions. The framework models interactions between phishing agents and three user personas representing cautious, average, and impulsive behavioral profiles, enabling comparative analysis of static and adaptive attack strategies. The framework was evaluated through 1,800 simulated interaction instances generated within a controlled simulation environment across multiple attack scenarios. Performance was measured using engagement-oriented metrics, including precision, recall, and F1-score. Experimental results indicate that adaptive strategies achieve an average improvement of 12-18% in F1-score compared to static approaches, demonstrating stronger alignment with user context across repeated interactions. By modeling phishing as an evolving behavioral process rather than a single event, the proposed framework provides a controlled and privacy-preserving environment for studying emerging AI-driven social engineering techniques. The approach enables quantitative analysis of adaptive phishing dynamics and supports the development of more effective defensive strategies against next-generation social engineering threats
Language: English
Topic: Spam and Phishing Detection
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7202316656