AI Fraud Detection: Agent Uncovers Hidden Schemes

An AI agent uncovered three significant findings after investigating fraud for eight hours without supervision. The agent, using a local vLLM serving Nemotron-3 Nano Omni and a cloud-hosted GLM-5.1, analyzed expense reports and six years of company data. What's interesting is that these findings were missed by a routine human review and even by the human who initially read the same documents. The agent identified an anomalous statistical pattern across six months of transactions, a phantom vendor with proper paperwork but few real company markers, and coordinated lunch receipts that revealed a different story when viewed together. The bottom line is that these long-horizon AI agents can keep thousands of data points active and cross-reference datasets in ways human auditors typically would not. This technology offers a patient, tireless, and holistic approach to forensic audit work, even running on local hardware. This capability matters for businesses of all sizes, from Fortune 500 companies to five-person startups, especially for fraud detection and compliance work.