Abstract

The financial services industry faces escalating fraud threats that exploit transactional complexity, distributed systems, and rapid digital adoption. Traditional rule-based fraud detection systems fail to keep pace with dynamic attack patterns and increasingly sophisticated adversarial strategies. This research investigates the integration of Artificial Intelligence (AI) and Large Language Model (LLM)-based cloud cybersecurity to enhance real-time fraud detection and prevention within financial institutions using scalable Extract, Transform, Load (ETL) workflows. The study proposes a hybrid architecture combining machine learning (ML) classifiers, natural language understanding (NLU), and LLM-enhanced anomaly detection layered with cloud-native security controls. A robust ETL pipeline is designed to process heterogeneous financial data streams—transaction logs, customer metadata, behavioral signals—securely within cloud infrastructure. We explore how LLMs improve contextual threat detection by interpreting semantic patterns in transaction narratives and user communication, thereby enhancing risk scoring. The research methodology involves data preparation, model training, cross-validation, and deployment within a cloud ecosystem equipped with AI-driven cybersecurity orchestration. Experimental results demonstrate improved detection accuracy, reduced false positives, and scalable performance under high transactional loads. The study concludes that aligning AI, LLM capabilities, and cloud cybersecurity with scalable ETL frameworks significantly advances financial fraud mitigation, offering practical implications for industry adoption.