China Investigates DeepSeek and Moonshot AI Over Data Transfers to Claude

China's internet regulator is investigating DeepSeek and Moonshot AI after Anthropic stated that both companies redirected users' sensitive data to Claude's models. The Information reports this citing its sources.

Anthropic's 154-page report, published on September 10, describes how seven Chinese companies used Claude on a large scale and without the company's permission. According to the report, DeepSeek was sending Claude requests from engineers who were working on a police surveillance system in China. Anthropic's report also notes that some sessions involving DeepSeek's and other Chinese companies' use of Claude contained sensitive information related to users, companies, and government-affiliated organizations.

Following the report's publication, China's Cyberspace Administration (CAC) summoned representatives of the seven companies named in the document for questioning. These included Alibaba, Zhipu, SenseTime, MiniMax, and Xiaomi, as well as DeepSeek and Moonshot AI.

According to Anthropic, DeepSeek conducted more than 12.1 million exchanges with Claude's models in just 14 days in July 2026. The company estimates that DeepSeek was automatically redirecting some users to Claude without their knowledge. One of the described cases involves engineers working on a police system intended for a Chinese municipal public security bureau, whose requests were being passed to Claude.

Anthropic's report also states that Moonshot AI conducted more than 23 million exchanges with Claude's models from May to July. The company claims that Chinese AI labs used Claude's responses in a so-called distillation process to improve the capabilities of their own models.

This case is significant not only because Anthropic detailed the unauthorized use of its own models in the report, but also because it provided specific examples in which information passed to Claude may have contained government, police, and other sensitive data. China's position on the specific investigation has not yet been fully presented at this stage.

According to Anthropic's report, the company identified and blocked the relevant accounts and strengthened its systems for detecting similar activity.