China’s AI race with US could turn its own data into strategic asset

China’s efforts to close the gap with the United States in artificial intelligence may carry an unexpected cost: in seeking to distil the capabilities of leading US AI models, Chinese companies could be sending vast amounts of sensitive user and industrial data back to the United States, according to an analysis published by Foreign Affairs.

The analysis examines a growing but largely overlooked dimension of the US-China AI competition: the strategic value of the data generated when people use AI for real-world work.

A recent report by US AI company Anthropic accused several Chinese AI firms, including Moonshot AI and DeepSeek, of conducting large-scale distillation of Claude’s capabilities. Anthropic said some companies went beyond generating training data and, in one case, routed at least 300,000 requests from real users through the model over a 10-day period.

According to the Foreign Affairs analysis, the material reportedly sent to Claude included surveillance material, internal code and valid credentials from engineers at major Chinese state-owned enterprises, police case-management information and details of internal AI projects at Chinese technology companies.

But the analysis argues that the larger strategic concern may not be individual pieces of sensitive information reaching the United States. Instead, it could be the cumulative value of vast quantities of real Chinese user data collected while companies attempt to extract capabilities from US models.

AI competition, the analysis argues, is increasingly about data as well as computing power. Publicly available internet data is becoming easier to obtain, while information generated through real tasks is much scarcer. AI interactions can reveal why users seek help, what problems they are solving, what they have already tried, which technical obstacles they face and what they intend to do next.

Individually, such interactions may appear insignificant. But aggregated across hundreds of thousands or millions of users, they could reveal broader patterns in China’s technological and industrial systems.

A programmer asking an AI model why a piece of code failed may reveal little. Thousands of engineers repeatedly reporting similar problems, however, could expose technological bottlenecks. Likewise, interactions involving workers in semiconductors, robotics, aerospace, telecommunications, electricity and automobile industries could gradually create a detailed picture of China’s industrial capabilities.

The Foreign Affairs analysis describes this as “intent data” — information that can reveal not simply what China possesses, but what it is doing.

That could become particularly valuable as tensions between Washington and Beijing increase. Combined with satellite imagery, procurement records, patents, corporate information and other intelligence, large volumes of AI interactions could potentially help identify emerging production bottlenecks, technological weaknesses and shifts in industrial activity.

The analysis raises a further question: whether Beijing previously understood that Chinese AI companies were potentially transferring real user data to US models while distilling their capabilities.

China already has extensive regulations governing data security and cross-border data transfers. But the analysis suggests that its regulatory system may not have fully anticipated a situation in which companies effectively export data through AI model calls rather than conventional databases.

This leaves Beijing with a difficult trade-off. Cutting off access to US frontier models could protect Chinese data but make it harder and more expensive for Chinese companies to acquire advanced AI capabilities. Allowing large-scale distillation, meanwhile, could allow valuable Chinese data to continue flowing into US systems.

The Foreign Affairs analysis concludes that as AI competition increasingly revolves around distillation, data and model capabilities, China faces a fundamental dilemma: how to benefit from the most advanced US AI technology without allowing its own data to become a strategic asset for the United States.

By Sabina Mammadli