DeXposure-Claw: An Agentic System for DeFi Risk Supervision

Researchers introduce DeXposure-Claw, an AI-powered supervision system designed to monitor DeFi credit risks by combining graph time-series forecasting with structured evidence gates to reduce false alarms in regulatory decision-making. The system includes a new evaluation benchmark aligned with regulatory standards, validated on five years of real blockchain data.

DeXposure-Claw addresses a critical gap in DeFi supervision: the tension between speed and accuracy in risk detection. Traditional LLM agents tend to over-escalate alerts based on weak signals, creating alert fatigue and undermining regulatory credibility. This system separates the signal-detection layer (DeXposure-FM forecasting exposure networks) from the decision layer (deterministic monitors and stress scenarios), preventing pure language models from driving high-stakes supervisory actions without grounding in quantitative evidence.

The research reflects growing recognition that DeFi's interconnected risk landscape—where credit cascades across protocols in real-time—requires fundamentally different supervision tools than traditional finance. As DeFi protocols manage billions in total value locked and systemic importance increases, regulators face pressure to intervene faster while maintaining evidential rigor. The introduction of DeXposure-Bench, a six-axis evaluation framework with explicit false-intervention rate metrics, provides the first regulator-aligned measurement standard for such systems.

For practitioners, this work signals maturation in the intersection of AI and regulatory technology. The system's success on historical data suggests scalable deployment is feasible, potentially accelerating adoption by blockchain monitoring platforms and regulatory agencies. The open-source release democratizes access to the methodology, enabling protocol teams and independent monitors to implement similar frameworks. However, the system's effectiveness depends on data quality and the accuracy of underlying forecasting models—weaknesses that remain to be tested under novel market conditions.

• →DeXposure-Claw combines LLM agents with deterministic monitors to reduce false-alarm escalations in DeFi risk supervision.
• →DeXposure-FM, a graph time-series foundation model, forecasts future exposure networks to ground supervisory decisions in quantitative evidence.
• →DeXposure-Bench introduces the first regulator-aligned evaluation framework with explicit false-intervention rate measurement.
• →System validated on five years of real blockchain data with open-source code released for broader adoption.
• →Architecture separates signal detection from decision escalation, preventing language models from driving high-stakes regulatory actions alone.