China’s internet regulator is investigating DeepSeek and Moonshot AI after allegations that the companies routed sensitive user data to Anthropic’s Claude models, The Information reports, citing people with knowledge of the matter. The outlet says the reported inquiry is examining whether data from Chinese military, police and state-owned corporate users was sent to the United States.
The status of the inquiry has not been independently confirmed in the public material reviewed here. Nor does the report establish a final finding against either company. The immediate issue is a set of allegations Anthropic made in its September threat-intelligence report, which describes what it says were unauthorized efforts to use Claude and extract its reasoning capabilities.
What Anthropic says it found
In its report, Anthropic says Moonshot, the company behind Kimi, silently forwarded some customer requests to Claude instead of processing them with Kimi. Anthropic says it observed almost 300,000 requests relayed during a ten-day period and that Moonshot retained at least some of the exchanges while trying to extract reasoning traces for model training.
Anthropic makes a similar allegation about DeepSeek. It says DeepSeek identified some users working through third-party or Anthropic coding tools and relayed selected requests to Claude Opus. The report says the campaign generated more than 12.1 million exchanges over 14 days in July. Anthropic describes examples involving sensitive corporate material and government-related data, while saying the customers whose requests were relayed were likely unaware of it.
Those are Anthropic’s findings and attributions, rather than independently established facts. Its report also says it does not know whether Moonshot notified affected customers. The public reporting reviewed here does not include a response from either company or a public statement from the Cyberspace Administration of China.
Why the alleged routing matters
The allegation is not simply that one model maker used another’s API. Anthropic says users believed they were sending prompts to a Chinese model but, in the cases it identified, some requests were sent onward to Claude without their knowledge. If that account is borne out, the privacy question is about the data in those prompts as well as the use of Claude’s output for training.
For now, the most precise description is a reported regulatory inquiry into allegations made by Anthropic. Whether China’s regulator has opened a formal case, what information it has sought, and how DeepSeek and Moonshot answer the claims remain unresolved.
