US data brokers are selling half a billion dollars in RL training environments to Chinese AI labs

Why it matters

A ChinaTalk investigation found that multiple American data firms, including Mercor, Surge AI, AfterQuery, and Turing, are selling reinforcement learning environments — simulated settings used to train models on tasks such as coding, financial analysis, or office work — to Chinese AI labs. Chinese buyers deliberately purchase the same products previously supplied to OpenAI or Anthropic, ensuring the data is proven and letting them skip expensive in-house development, a saving that matters all the more given memory bottlenecks in Chinese chips. The report says these relationships surfaced through anonymous sources and subcontractor sloppiness, since outsourced workers in developing countries sometimes leaked client names. Because such environments are largely unregulated, the authors urge the Bureau of Industry and Security to apply export controls to dual-use coding and biology environments, which could sharpen offensive cyber or bioweapons-relevant capabilities. They caution, however, that the situation is not apocalyptic: Chinese labs still trail American frontier companies by months.