Abstract

The increasing trust placed in cloud systems to store sensitive data has exposed a critical flaw in traditional security models, in which reactive methods are not sufficient to prevent data from being secretly taken. This paper provides an overview of the state-of-the-art in cloud security and proposes a forward-thinking solution. Current defenses only attempt to prevent leaks after they have begun, and this is too late — meaning a malicious individual, such as an untrustworthy employee, can steal valuable data before the security system even sends an alert. Our proposed method changes this by continuously checking what the user is doing. It relies on intelligent computer models to anticipate when an unusual or potentially risky situation may be occurring and provides time for the system to respond automatically. The basis of this system is a robust, three-factor authentication mechanism (using a combination of static passwords, advanced voice biometrics based on ECAPA-TDNN, and One-Time Passwords), built into an Artificial Intelligence (AI) engine. This engine uses deep learning models (LSTM/Autoencoders) to predict abnormal user behavior and prevent data leaks from occurring, performing continuous trust verification instead of a single login check. By analyzing 50 recent research contributions (2020-2025), we illustrate the need for moving past disjointed security solutions toward a comprehensive and continuously verified solution for the secure storage of sensitive cloud vaults.