Practical Visual Detection of Social Engineering Attack Campaigns

Social engineering attacks threaten users’ and organizations’ security andprivacy. Existing solutions for social engineering attacks are limited in scope and reactiveness, lacking a comprehensive approach to tackle the fundamental traits of these attacks and address emerging threats. We present SEShield, a framework to detect web-based social engineering attacks beyond phishing to fill this gap. SEShield consists of three components: SECrawler for identifying new attack campaigns; SENet, a fully convolutional neural network utilizing federated learning for visual detection; and SEGuard, a real-time browser extension for classification and user alerts. The research primarily evaluated SENet, demonstrating its effectiveness in detecting new instances of social engineering attacks. SENet yielded a detection rate of 99.6% at 1% false positive rate when tested on new SE instances. The proposed framework strengthens defenses against generic social engineering attacks.