New benchmark tests AI agents' ability to detect fraud and manage risks in real e-commerce environments with 1,513 production tasks.

RiskWebWorld is the first interactive benchmark specifically designed to evaluate GUI agents in high-stakes e-commerce risk management scenarios

Includes 1,513 tasks from production risk-control pipelines across 8 core domains, capturing authentic challenges on uncooperative websites

Features Gymnasium-compliant infrastructure that separates policy planning from environment mechanics to support scalable evaluation and reinforcement learning

Evaluation reveals significant capability gaps between different AI models in handling real-world risk operations and environmental hijackments