China investigates DeepSeek and Moonshot for leaking data to Anthropic

China's internet regulator, the Cyberspace Administration, is investigating two of the country's big AI companies, DeepSeek and Moonshot. The background is an accusation from the American AI company Anthropic, which makes the chatbot Claude.

On September 10, Anthropic published a 154-page report in which the company accuses seven Chinese AI companies of secretly passing users' requests on to Claude. According to Anthropic, customers believed they were using a Chinese product, while their data was in reality being processed by the American Claude system, partly to harvest answers for training their own models.

According to Anthropic, Moonshot passed more than 23 million requests on to Claude between May and July. DeepSeek allegedly did the same with more than 12 million requests in just 14 days in July.

What particularly worries the Chinese authorities is what type of data was potentially sent abroad. According to the information, DeepSeek allegedly forwarded data that could be linked to police records and national ID numbers, among other things. Moonshot allegedly sent surveillance images from several hundred cameras in the city of Chengdu, including cameras near facilities linked to the Chinese military and state-owned companies.

Chinese authorities have summoned representatives from all seven companies Anthropic names in its report but have so far chosen to focus the investigation on DeepSeek and Moonshot. The case is still open, and no decision has yet been made on possible sanctions.

The case comes at a sensitive time in relations between the US and China. It is taking place while negotiations are underway on export rules for advanced AI chips, and shortly before a planned meeting between the American and Chinese governments. For ordinary users of AI services, the case underlines that it is not always transparent which company is actually processing your data when you use a chatbot.