Explainable AI for Cyberscam Prevention: A Survey of Multimodal Scam Detection, Evaluation, and Human-Centered Training
Cyberscams increasingly exploit social engineering, synthetic media, and human cognitive vulnerabilities, yet most Artificial Intelligence (AI)-based detection systems remain difficult for users to understand and act upon. Explainable AI (XAI) has emerged as a promising approach for improving transparency in scam detection, but its role in supporting user learning and scam prevention remains unclear. This survey reviews XAI for cyberscam detection and prevention across phishing, spam, deepfakes, voice scams, and multimodal deception.
We synthesize research along three dimensions: scam modality, explanation technique, and application perspective, covering datasets, benchmarks, and evaluation practices. Beyond model-centered explainability, we introduce an educational perspective grounded in learner knowledge dimensions and examine how explanations can support scam awareness, understanding of deception strategies, verification behaviors, and reflective judgment. Our analysis reveals a persistent gap between explanation generation and human impact, as most studies emphasize technical performance while rarely evaluating learning, trust calibration, behavioral outcomes, or scam resilience. We conclude by identifying key challenges and future directions for human centered, learning-centered, and scam-journey-oriented XAI systems that better support cybersecurity education and scam prevention.
Publications
Under Review (ACM CSUR)
Authors: Heajun An, Hossein Salemi, Qi Zhang, Hemant Purohit and Jin-Hee Cho