Agentic AI in Cloud-Based Credit Card Fraud Detection: Towards Autonomous Risk Mitigation

Abstract

Credit card fraud (CCF) still remains an ongoing concern for financial institutions because to the huge disparity between genuine and fraudulent transactions, as well as the ever-changing behavior of fraudsters. This paper proposes a cloud-based Agentic Artificial Intelligence system for real-time credit card fraud detection that utilizes autonomous multi-agent collaboration and deep temporal modeling. The system makes use of the publicly accessible CCF Detection dataset, commencing with secure cloud-level data ingestion, preprocessing, and normalization.  An agentic transaction analysis layer made up of qualified agents accomplishes data validation, behavioral pattern analysis, and transaction history verification. To successfully capture both local spatial aspects and long-term temporal dependencies in transactional behavior, deep temporal fraud modeling employs a hybrid full-dimensional dynamic convolutional network mixed with shuffle attention and LSTM. Finally, an automated risk assessment and decision module generates fraud scores and initiates relevant mitigation steps such as alarms or transaction blocks. Experimental results show that the proposed framework outperforms baseline GRU-based models in detection performance, demonstrating its effectiveness, scalability, and appropriateness for real-time intelligent fraud prevention in cloud environments.