Chinese grey market sells Claude API access at 90% off by using stolen credentials, model substitution, and harvesting users' prompts and outputs for resale as AI training data

A recent investigation by Zilan Qian of the Oxford China Policy Lab reveals a thriving grey market in China that resells access to Anthropic’s Claude models at prices as low as 10% of the official rat... Read original →

This gray market highlights critical vulnerabilities in AI commercialization—particularly around model security and data governance. High pricing of closed-source models like Claude creates arbitrage opportunities, driving actors to exploit stolen credentials and deceptive model substitution (e.g., serving Qwen while labeling it as Claude). Simultaneously, user prompts and outputs are harvested and resold as training data, risking data contamination and IP leakage. From a competitive standpoint, firms like Anthropic face not only revenue erosion but also reputational damage from inconsistent or misrepresented model performance. The modular, adaptive nature of these proxy networks underscores the inadequacy of current verification systems. Without robust solutions—such as behavioral biometrics, output watermarking, and auditable data lineage—the AI industry risks systemic trust degradation. This could accelerate regulatory fragmentation, especially in markets like China, where authorities may impose stricter data localization rules, ultimately reshaping global AI deployment strategies.