Credit card activation and registration flows operating in non-logged-in environments represent one of the most attractive targets for sophisticated automated attacks in the financial services sector. The absence of prior authentication, combined with the high value of successfully activated accounts, makes these flows particularly susceptible to low-and-slow card testing campaigns, enumeration attacks, and synthetic identity creation attempts. Traditional defensive mechanisms, including static device fingerprinting and rule-based rate limiting, have proven increasingly inadequate against modern evasion techniques that leverage browser automation frameworks, anti-detect browsers, and AI-generated behavioral patterns. This paper presents a comprehensive edge-deployed artificial intelligence and machine learning framework designed to detect and preempt such threats at the earliest possible stage in the request lifecycle. By collecting and analyzing multi-modal signals encompassing behavioral biometrics, temporal dynamics, and network characteristics directly at the content delivery network and web application firewall layer, the proposed system constructs rich risk profiles in real time. An ensemble of unsupervised anomaly detection models and supervised classifiers operates on over 120 engineered features to produce calibrated risk scores, enabling immediate mitigation decisions—allow, silent challenge, or drop—without any request ever reaching the origin application servers or invoking backend application programming interfaces.