As smartwatches become increasingly integrated into daily life, their electromagnetic (EM) emissions introduce a significant yet overlooked privacy risk. This study systematically examines how EM leakage from smartwatches can be exploited to infer user interactions and behavioral patterns. We propose MagWatch, a novel non-intrusive attack that applies wavelet transform for signal processing and leverages a CNN-LSTM model to identify applications and in-app activities, achieving up to 90% accuracy across multiple smartwatch models. Our findings reveal a critical security vulnerability, demonstrating that attackers can passively monitor EM emissions to reconstruct user interactions, exposing sensitive information such as communication habits and app usage patterns. This research highlights the urgent need for privacy-preserving countermeasures in wearable technology and establishes a foundation for future studies on EM side-channel security risks.
