Sources: the Cyberspace Administration of China is investigating DeepSeek and Moonshot over alleged data leaks to Anthropic via Claude and has questioned staff
China's internet regulator is investigating DeepSeek and Moonshot AI after Anthropic alleged both companies had been routing sensitive user data …
Context & Ripple Effects
The reported CAC inquiry follows Anthropic's February allegation that DeepSeek, MiniMax and Moonshot made more than 16 million Claude prompts for distillation, and its September account of DeepSeek and Moonshot using overseas “transfer stations” to send queries to Claude. Anthropic's claims remain allegations, but a domestic regulator examining the purported routing turns a terms-of-service dispute into a compliance issue for the two Chinese labs.
First-order effects
- DeepSeek and Moonshot face reported CAC scrutiny and staff questioning over whether data passed through the alleged Claude-routing channels, placing their handling of user prompts under regulatory review.
- Anthropic's allegations gain a second consequence beyond its own platform rules: the reported inquiry focuses attention on whether its model access exposed sensitive data.
Second-order effects
- Chinese AI companies named in Anthropic's earlier claims have reason to audit external model-access intermediaries and the data flows they create, because the CAC reportedly summoned all seven companies referenced in the report.
- Providers and operators of cross-border routing infrastructure face greater compliance risk when they sit between Chinese AI labs and foreign models, even where the underlying use is framed as model evaluation or distillation.
Third-order effects
- If regulators treat foreign-model routing as both a data-governance and industrial-policy issue, Chinese AI labs will be pushed toward more self-contained training and inference arrangements rather than dependence on external frontier models.
- The episode points to model access becoming a state-mediated control point: platform terms, data rules and national AI policy can jointly determine which labs can use foreign models and through what infrastructure.
The trend: Cross-border access to frontier models is becoming governed not only by provider restrictions but also by national regulators policing the data and intermediaries around that access.