I Gave an AI Agent 8 Hours to Investigate Fraud. Here's What It Found

This article explores the design of a long-horizon AI agent built for forensic expense auditing and fraud detection. Using local LLM infrastructure, sandboxed execution environments, structured policy enforcement, and automated document workflows, the system analyzes expense reports for anomalies, suspicious vendors, policy violations, and coordinated fraud patterns without relying on third-party cloud APIs.