China's Cyberspace Administration is investigating DeepSeek and Moonshot AI after Anthropic alleged that user data from China may have been routed to its Claude model without users' knowledge. Anthropic's 154-page report accused seven China-based labs of "illicit distillation," estimating about 190 million routed exchanges across the group, with Moonshot and DeepSeek accounting for millions of those. The probe complicates near-term plans for both firms, including a U.N. briefing and a planned Hong Kong IPO, and raises potential Chinese data-security concerns.
China Probes DeepSeek and Moonshot After Anthropic Says Millions of User Exchanges Were Routed to Claude
China is investigating DeepSeek and Moonshot over data that flowed to Anthropic hero image
China's Cyberspace Administration (CAC) has opened an investigation into AI firms DeepSeek and Moonshot AI after reports that Chinese users' data may have been transmitted to Anthropic's Claude model without those users' knowledge, The Information reports. Officials have visited both companies to question executives and staff; the inquiry is ongoing and no decisions on penalties have been announced.
The probe followed a 154-page threat intelligence report Anthropic published on Sept. 10 that accused seven China-based AI labs of routing customer requests through Anthropic’s Claude model to harvest outputs for model training — a practice Anthropic described as "illicit distillation." The firms named were Alibaba, Moonshot AI, DeepSeek, Zhipu, MiniMax, Xiaomi and SenseTime. After the report surfaced, Chinese regulators summoned representatives from all seven firms before focusing the inquiry on DeepSeek and Moonshot, according to The Information.
Regulatory Focus Versus Anthropic’s Complaint
Anthropic's complaint centers on Chinese labs siphoning capability from its model. Beijing’s concern runs the other way: that personal or sensitive data belonging to Chinese users ended up on servers of an American company without proper notification or legal safeguards, potentially breaching Chinese data-security laws.
Specific Allegations
Anthropic’s report includes detailed examples involving both companies. In one case, it says a user linked to the People’s Liberation Army asked Moonshot’s chatbot "Kimi" to analyze surveillance footage tracking a single person across hundreds of police cameras in Chengdu, including cameras outside PLA facilities and defense-linked institutions; Anthropic alleges Moonshot forwarded the request and the footage to Claude without informing the user. In another example, Anthropic alleges DeepSeek relayed queries from engineers building a police surveillance case-management system, exposing live credentials tied to a Russian government database.
Scale of Activity
Anthropic estimates the combined "distillation" activity across the seven named firms at roughly 190 million exchanges. It says Moonshot routed more than 23 million exchanges through Claude between May and July, while DeepSeek generated over 12.1 million exchanges in a 14-day period in July. Despite Alibaba accounting for more than 151 million of the exchanges, Alibaba is not currently the focus of the CAC inquiry.
Timing and Potential Consequences
The inquiry has immediate commercial and diplomatic implications. DeepSeek is scheduled to brief the U.N. Security Council this week on AI risks at the U.N. General Assembly, and Moonshot has confidentially filed for a Hong Kong IPO targeting about $3 billion — a regulatory probe would likely need to be disclosed in any prospectus. The investigation also arrives as President Xi Jinping prepares to visit Washington, where AI and mechanisms for cross-border notification of AI incidents are expected to be raised in talks with President Donald Trump.
What’s Next
The CAC has not announced a timetable or indicated whether either company will face penalties. The probe highlights growing scrutiny from regulators worldwide over how AI systems route and handle user data and the legal and national-security questions that can follow.
Help us improve.
