China's Grey Market for Banned US AI Models Thrives via Transfer Stations
Chinese developers are accessing restricted U.S. AI models at significantly lower costs through "transfer stations," API proxy services. This grey market operates on platforms like GitHub and Taobao, exceeding initial U.S. government assessments of its scale.
A substantial "grey market" has emerged in China, enabling developers to access U.S. artificial intelligence models that are officially restricted. At the heart of this ecosystem are "transfer stations," which function as API proxy services. These intermediaries allow Chinese developers to use advanced AI models, including those from Anthropic, for as little as 10% of the official price.
U.S. authorities and AI companies like Anthropic have implemented measures such as geo-blocking, phone number verification, and credit card requirements to prevent access from China. Anthropic recently introduced mandatory live selfie and government ID verification for select users. However, these efforts have spurred the development of sophisticated evasion infrastructure, including the proliferation of SMS farms and biometric data harvesting operations.
The user base for these services extends beyond experienced AI researchers to include university faculty, students, tech workers, individual developers, and hobbyists. The data logs generated through these proxies are commoditized and used for various purposes, including model training and targeted fraud. The "proxy" middlemen are deliberately designed by some Chinese labs to systematically extract U.S. AI models.
This "transfer station" economy demonstrates the limitations of access blocking and monitoring as AI governance tools. It highlights blind spots in safety frameworks intended to prevent misuse by malicious actors and protect provider traceability, operating beyond the U.S.-China rivalry.