The rapid digitalization of banking services has significantly increased the exposure of financial systems to identity-based fraud, necessitating advanced and adaptive authentication mechanisms. This study proposes an AI-driven facial recognition framework leveraging Google Teachable Machine to enhance fraud detection capabilities in banking security systems. The proposed framework integrates lightweight deep learning-based facial feature extraction with browser-accessible machine learning tools to enable real-time, scalable, and cost-effective biometric authentication. The research critically examines how facial biometrics, when combined with user-friendly AI training environments, can reduce dependency on traditional password-based systems and improve fraud resilience.

A structured methodological approach is adopted, combining dataset preparation, model training using Teachable Machine, and performance evaluation under simulated banking authentication scenarios. The study draws insights from biometric authentication literature, fraud detection systems, and hybrid machine learning models to design a robust conceptual architecture. Findings indicate that AI-driven facial recognition systems can significantly reduce unauthorized access risks, particularly when combined with anti-spoofing mechanisms and behavioral validation layers. However, challenges such as dataset bias, environmental variability, and presentation attacks remain critical limitations.

The study concludes that integrating accessible AI tools like Google Teachable Machine into banking security frameworks can democratize biometric system deployment while enhancing fraud detection efficiency. The research contributes a scalable architectural model suitable for next-generation digital banking environments.