China investigates whether DeepSeek and Moonshot routed sensitive user data through Claude

China’s internet regulator is investigating DeepSeek and Moonshot AI after Anthropic alleged that both companies routed sensitive user data to Claude models. The inquiry follows Anthropic’s 154-page report of September 10, which described seven Chinese companies using Claude at scale despite access restrictions.

The allegations should remain distinct from established findings: Anthropic produced the underlying account, while the Chinese investigation will determine what data moved, under whose instructions and whether domestic privacy or security rules were breached. Neither a model request nor a foreign API call automatically proves that protected user information was transferred.

The episode exposes the conflict between model sovereignty and practical capability. A Chinese company may advertise a domestic model while using a stronger foreign model behind a tool, evaluator or workflow. That can improve performance quickly, but it also creates undisclosed subprocessors, cross-border data flows and dependence on a supplier that can revoke access.

Beijing’s response is therefore about more than enforcement against two laboratories. It tests whether China can require provenance for model outputs and data paths while simultaneously pushing its AI sector to compete with American systems.