China Regulators Probe AI Firms over Data Routing
Chinese authorities are investigating two major AI companies after allegations that user data was sent to a US model for training purposes.
Key points
- China's Cyberspace Administration is probing DeepSeek and Moonshot AI over alleged data routing to Anthropic's Claude model.
- Anthropic claims the firms sent millions of user prompts, including sensitive personal and corporate data, to use for training their own systems.
- The investigation aligns with upcoming US-China diplomatic talks, potentially signaling China's commitment to stricter AI regulation and data privacy standards.
China’s internet regulator has opened an investigation into DeepSeek and Moonshot AI, two prominent domestic artificial intelligence developers. The probe follows serious allegations that these companies routed user prompts and sensitive data to Anthropic’s Claude model, a competitor based in the United States.
According to reports from Gizmodo, the Cyberspace Administration of China has summoned executives from several leading AI labs for questioning. While the initial scrutiny was broad, regulators have since narrowed their focus specifically to DeepSeek and Moonshot AI, citing potential violations of privacy laws and service terms.
Accusations of Data Distillation
The controversy stems from a detailed report published by Anthropic, which accused the Chinese firms of using a technique known as distillation. This process involves sending massive amounts of user inputs to a more powerful external model and then using the generated responses to train smaller, in-house systems. Anthropic claims this allowed the competitors to improve their outputs without building equally large training datasets from scratch.
The scale of the alleged data transfer is significant. Anthropic reported that Moonshot AI sent over 23 million user inputs to Claude between May and July, while DeepSeek transmitted 12.1 million prompts during a two-week period in July. Critics argue this practice effectively shortcuts the intensive work required to develop robust AI capabilities, raising ethical questions about fair competition in the global tech market.
Privacy Risks for End Users
Beyond the competitive implications, the investigation highlights serious privacy concerns. Anthropic stated that the routed prompts contained sensitive information from individual users, multinational corporations, and state-affiliated actors. This data reportedly included names, email addresses, and proprietary company details in at least a dozen different languages.
Chinese regulators appear concerned that such practices may violate local privacy statutes and the companies' own terms of service. By exposing user data to a third-party American model, the firms risked compromising the confidentiality of information belonging to hundreds of end users. This legal exposure likely forms a central part of the current regulatory scrutiny.
Strategic Timing of Inquiry
The timing of this probe is notable, as it coincides with upcoming diplomatic talks between US President Donald Trump and Chinese President Xi Jinping. Artificial intelligence is expected to be a key topic in these discussions, with both sides seeking a framework for global regulation.
Analysts suggest that China’s move to investigate its own firms may serve as a diplomatic gesture. By demonstrating active oversight and enforcement, Beijing aims to show it is operating in good faith regarding data security and AI standards. This could help mitigate tensions ahead of high-level negotiations on the future of technology governance.
