A very significant risk and complexity which has emerged in fraudulent activities in banking transactions is the accelerated rate at which the financial services have become digitized. This study brings forth a state-of-the-art AI enabled fraud detection framework utilizing deep learning technology to increase the accuracy and real-time response of detecting anomalous patterns. More specifically, we deploy a Long Short-Term Memory (LSTM) neural network architecture because of its enhanced ability to learn the temporal dependencies in sequential transaction data. A real world financial dataset is used to train and test the model using Google Cloud AI Platform that guarantees scalable processing and integrated model deployment. Through extensive experiments it is shown that the proposed system is capable of delivering high precision and recall when it comes to detecting fraudulent behavior and outperforms traditional ML baselines. Such research emphasizes the effectiveness of the deep learning technology in automating fraud detection and demonstrates the breakthrough novelty of cloud-based AI tools to the financial industry.