DeXposure-Claw: An Agentic System for DeFi Risk Supervision

2026-06-17

AI Summary

This study addresses the challenges of rapidly evolving credit risk in decentralized finance (DeFi) and the susceptibility of large language model agents to false alarms triggered by weak evidence. To this end, it proposes the first regulatory-aligned DeFi risk surveillance framework, which integrates a graph temporal foundation model (DeXposure-FM) to forecast risk exposure networks, deterministic monitoring rules, and stress-testing protocols to generate structured alerts with attributable evidence. The framework further incorporates data health assessments and confidence-gated mechanisms to ensure auditable, regulator-ready outputs. Evaluated on five years of real-world weekly data using a novel six-dimensional assessment protocol, the approach significantly reduces both erroneous interventions and absolute losses. The work also contributes the open-source DeXposure-Bench benchmark and associated codebase to foster reproducible research in DeFi risk monitoring.

Abstract

Decentralized finance exposes supervisors to fast-moving, networked credit risks. General-purpose LLM agents fit this setting poorly: they over-read weak evidence and recommend high-stakes interventions, while existing evaluations offer no regulator-aligned way to measure the resulting false alarms. We introduce DeXposure-Claw, a forecast-grounded agentic supervision system that routes LLM decisions through structured evidence: (1) DeXposure-FM, a graph time-series foundation model, forecasts future exposure networks; (2) deterministic monitors and stress scenarios then turn those forecasts into typed alerts, attribution signals, and scenario evidence; and (3) data-health and confidence gates constrain escalation before DeXposure-Claw emits auditable supervisory tickets with rationales. We further develop 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 fully support our system. Code is at https://github.com/EVIEHub/DeXposure-Claw.

Problem

• DeFi risk supervision
• credit risk
• false alarms
• regulator-aligned evaluation
• LLM agents

Innovation

• agentic supervision
• graph time-series foundation model
• structured evidence
• false-intervention rate
• DeFi risk monitoring