Smart wearable devices, including smart watches, fitness trackers, augmented reality (AR) headsets, and biometric sensors, are widely used across healthcare, finance, and enterprises. However, their reliance on wireless connectivity (Wi-Fi, Bluetooth, and NFC) makes them vulnerable to cybersecurity threats such as man-in-the-middle (MITM) attacks, unauthorized access, and data breaches. Traditional security solutions, including signature-based intrusion detection and centralized machine learning models, face limitations due to high computational costs, privacy concerns, and scalability issues. Federated Learning (FL) takes a different direction than a traditional centralized security model. FL fundamentally allows wearable devices to learn and train on the local wearable networks intrusion detection models that protect data and reduce the possibility of failure of a centralized attack. There is a real-world implementation of the system using the Aegean Wi-Fi Intrusion Dataset (AWID), which indicates and identifies several types of cyber-attacks on Wi-Fi connected wearable networks. The experiments show that FL smart autoencoders can meet the detection objective while demonstrating high accuracy, low false positive rates and receiver operating characteristics (ROC) regulatory compliance detection. The key point of this paper is to explain how FL-based AI models can be considered the next generation of smart IoT wearables by providing privacy-reserving adaptable scalable security architectures.