LLM-Powered Agent for Fraud Report Triage
A new AI agent uses large language models to triage bank fraud reports through multi-turn conversations, probing for details and routing cases to specialist teams. The system simulates synthetic digital twins of real customers to generate realistic dialogues for evaluation and improvement. It integrates policy, safety guardrails, and reasoning frameworks to handle millions of annual fraud, scam, and dispute reports, aiming to replace slow manual processes that cause stress for customers and staff.
Key facts
- arXiv:2605.16268v1
- Announce Type: cross
- Banks receive millions of reports of fraud, scams, and disputed transactions every year
- Existing manual process is slow and stressful
- AI agent uses LLMs for multi-turn conversations
- Agent asks relevant questions and classifies cases for policy-guided routing
- Synthetic digital twins of real customers are simulated
- Digital twins generate realistic, labelled dialogues based on historical data
Entities
Institutions
- arXiv