Cloud-Native Financial Cybersecurity: Deploying Serverless Large Language Models for Real-Time Fraud Detection and AI-Driven Scam Neutralization

The proliferation of Generative Artificial Intelligence has armed cyber adversaries with unprecedented capabilities to deploy highly sophisticated, at-scale social engineering attacks and financial scams. Traditional rule-based fraud detection systems, which rely on static transaction thresholds, are structurally incapable of parsing the semantic complexity of these AI-driven threats. This paper proposes a cloud-native architectural framework utilizing Amazon Web Services to construct a real-time, serverless cybersecurity pipeline for digital financial networks. By deploying asynchronous Python middleware integrated with localized Large Language Models and eXtreme Gradient Boosting classifiers, the proposed system programmatically ingests high-frequency transaction metadata and multi-modal communication vectors. The system dynamically calculates a Semantic Fraud Probability Score, instantly quarantining malicious transactions before settlement. Preliminary architectural evaluations demonstrate that decoupling natural language threat analysis from legacy banking infrastructure significantly accelerates scam neutralization, providing information systems researchers with a deterministic, highly scalable technological foundation for combating emerging digital financial threats.