China Probes AI Firms Over Data Leaks to Anthropic

The Cyberspace Administration of China is investigating seven domestic technology companies for allegedly routing sensitive user data to Claude AI models developed by the U.S. firm Anthropic. The probe focuses on violations of strict cross-border data rules, with reports indicating that companies used unauthorized access to interact with the U.S. models. Moonshot AI allegedly sent over 23 million interactions between May and July, while DeepSeek handled more than 12 million interactions during a 14-day period in July.

The leaked data reportedly included corporate secrets, police data, and information regarding People’s Liberation Army facilities. DeepSeek specifically escalated engineering requests involving law enforcement monitoring infrastructure to the U.S. models. Other companies summoned for questioning include Alibaba, Zhipu, SenseTime, MiniMax, and Xiaomi. Following the news, shares of Chinese AI developers declined in Hong Kong, with Z.AI and MiniMax Group falling as much as 7.4% and 6.3% respectively.

Simultaneously, Anthropic reported disrupting a Chinese-government-aligned operation that used Claude to gather intelligence on religious leaders and diaspora communities. The operation targeted the International Campaign for Tibet, Catholic leaders in Asia, and Falun Gong practitioners to create dossiers for potential exploitation. Anthropic has since terminated the accounts associated with these surveillance activities. These developments occur as investors await a summit between U.S. President Donald Trump and Chinese President Xi Jinping to discuss the AI industry.