The swiftly emerging field of Continuous Multimodal Biometric Authentication (CMBA) emphasizes the possibility of providing strong and real-time verification of a person’s identity through various means, including facial features, iris pattern, voice characteristics, and gait. A CMBA framework, which is Deep learning oriented, is presented in this paper that resolves issues related to fusion complexity, privacy, and dataset diversity. A layered architecture is proposed here, having phases. A pilot simulation was conducted to test the framework practically, where throughout the class, the users were continuously authenticated in class using four biometric modalities. Here, 97.5% of TMR was given, FAR was about 1.3%,98.2% was the spoof detection accuracy, and 1.8 s was the average authentication latency. When we compare this with the current state-of-the-art multimodal models, the proposed system performs better than them all, primarily in terms of spoofing as well as error reduction rate. The consequences would be that a technically sound and scalable CMBA can be achieved, which can be applicable for real-world applications for automatic attendance marking, which is privacy-conscious.