Abstract
Technological advances that provide for instant data availability also increase the possibilities for hackers to tamper with data causing cybercrime. To address this challenge, we propose a hybrid trust-enhancing AI framework that augments traditional ML predictions with reasoning from a large language model (LLM), specifically GPT-3.5. The framework allows not only for classification of transactions but also for natural language justification of each decision, making the model’s behavior more interpretable, auditable, and trustworthy.