The Cyberspace Administration of China is investigating AI firms DeepSeek and Moonshot for leaking sensitive domestic data to Anthropic. The probe examines whether the companies secretly routed customer requests to the U.S. AI lab's Claude models to harvest training data, potentially violating national security laws. Alleged leaks include People’s Liberation Army surveillance footage from Chengdu, police records, and credentials for a Russian government defense database.

DeepSeek routed 12.1 million exchanges in 14 days in July, while Moonshot forwarded over 23 million requests between May and July. Regulators initially questioned representatives from seven companies, including Alibaba, Xiaomi, SenseTime, and MiniMax, before narrowing the focus to DeepSeek and Moonshot. No penalties have been announced as the internet regulator reviews the cross-border transfer of state and corporate information.