AI-Driven Security for Financial Transactions: Leveraging LLMs, Federated Learning, and Behavioral Biometrics

The rising sophistication of cyber threats ranging from phishing and synthetic identities to adversarial model attacks has made the demand for intelligent, adaptive security solutions in financial systems more urgent than ever. This study explores key AI technologies that are shaping the future of secure financial transactions, including Large Language Models (LLMs), Federated Learning (FL), Graph Neural Networks (GNNs), and behavioral biometrics. For each technology, we outline its core architecture, operational mechanisms, and applicability in real-world fraud detection systems. LLMs enable contextual understanding of transaction narratives, aiding in the detection of phishing attempts across various communication channels. FL facilitates collaborative model training across multiple financial institutions without compromising user privacy. GNNs leverage the relational structure of transaction networks to uncover fraud rings that evade traditional rule-based systems. Behavioral biometrics offers continuous authentication by analyzing passive user attributes such as typing patterns and device interaction. A comparative analysis demonstrates the advantages of these AI approaches over conventional methods, highlighting improvements in detection accuracy, scalability, and privacy preservation. The review also addresses critical challenges including data imbalance, latency, model drift, and regulatory constraints. Together, these insights provide a comprehensive foundation for understanding how AI, when applied responsibly, can enhance the integrity and resilience of financial ecosystems