LLMs Were Overwhelming DeFi Risk Teams with False Alarms — Here’s the System That Finally Fixes It

Decentralized finance (DeFi) exposes regulators to fast-moving, networked credit risks that generic LLM agents handle poorly—they over-read weak evidence and recommend high-stakes interventions, leading to costly false alarms. To address this, researchers from multiple institutions introduce DeXposure-Claw, a forecast-grounded agentic system that routes LLM decisions through structured evidence. The system comprises three components: (1) DeXposure-FM, a graph time-series foundation model that forecasts future exposure networks; (2) deterministic monitors and stress scenarios that convert forecasts into typed alerts, attribution signals, and scenario evidence; and (3) data-health and confidence gates that constrain escalation before the system emits auditable supervisory tickets with rationales.

To measure performance, the team built DeXposure-Bench, a six-axis evaluation harness whose decision axis scores tickets against a regulator-aligned absolute-loss ground truth and an explicit false-intervention rate. Experiments on five years of weekly real data demonstrate that DeXposure-Claw significantly reduces false alarms compared to generic LLM agents, making it a practical tool for DeFi risk supervision. The code is publicly available. This work addresses a critical gap in applying LLMs to high-stakes financial regulation.