The Cyberspace Administration of China opened an investigation into artificial intelligence startups DeepSeek and Moonshot AI over allegations that both companies secretly routed user queries to Anthropic’s Claude models. Regulators launched the probe following a 154-page threat intelligence report published by Anthropic on September 10, which accused multiple Chinese labs of illicit model distillation.
Regulatory Scrutiny Targets DeepSeek and Moonshot AI
Chinese regulatory officials visited the offices of DeepSeek and Moonshot AI to interview executives and staff, according to reporting by The Information. The Cyberspace Administration of China initiated the inquiry to determine whether sensitive state-linked corporate, police, and military data ended up on United States servers during the query routing process.
The regulatory interest initially encompassed seven China-based labs named in the Anthropic threat intelligence report, including Alibaba, DeepSeek, Moonshot, Zhipu (Z.ai), MiniMax, SenseTime, and Xiaomi. Regulators subsequently narrowed their focus specifically to DeepSeek and Moonshot AI. No official penalties have been announced by Chinese authorities regarding the ongoing investigation.
Distillation Claims and Data Security Concerns
Anthropic’s September 10 report detailed what the American AI company classifies as illicit distillation. According to the report, Moonshot sent more than 23 million user queries to Claude between May and July, while DeepSeek routed 12.1 million requests to the model over a two-week period in July. The startups allegedly used responses generated by Claude to train and enhance their own smaller proprietary models, achieving higher efficiency while bypassing the extensive costs required to build large training datasets.
While American AI laboratories view the practice as unauthorized data extraction, Chinese state regulators are primarily focused on data security violations. Anthropic stated that the redirected queries contained highly sensitive information, including names, corporate data from international firms, email addresses, and details from structures linked to Chinese state authorities across at least 12 languages. Exposing this information to third-party foreign servers constitutes a direct violation of China’s strict data security and privacy laws.
High-Stakes Diplomatic and Financial Timing
The timing of the investigation coincides with major corporate and diplomatic milestones for both targeted firms. Moonshot AI filed confidentially for an initial public offering in Hong Kong on September 3, targeting a $3 billion raise at a $50 billion valuation. Under securities rules, any active regulatory investigation must be disclosed in the company’s prospectus before a public listing can proceed.
Simultaneously, DeepSeek is scheduled to brief the United Nations Security Council on artificial intelligence risks, sharing a platform with Anthropic CEO Dario Amodei. The United Nations session precedes the September 24 summit between United States President Donald Trump and Chinese President Xi Jinping, where artificial intelligence regulation and technological oversight are expected to form a central component of bilateral discussions.
Related reading