TL;DR
China's top internet regulator is investigating DeepSeek and Moonshot AI for allegedly routing tens of millions of user requests through Anthropic's Claude, potentially exposing sensitive Chinese military, police, and corporate data to a U.S. company.
What happened
- The Cyberspace Administration of China (CAC) sent officials to DeepSeek and Moonshot AI around September 22 to 23, 2026, interviewing employees about possible data security violations.
- Anthropic's September 2026 threat report alleged DeepSeek routed more than 12.1 million exchanges to Claude over 14 days in July, and Moonshot routed nearly 300,000 customer requests through 5,380 fraudulent accounts over ten days.
- Anthropic identified more than 23 million Moonshot exchanges from May through July, with Moonshot allegedly showing Claude responses to customers while presenting them as its own Kimi model outputs.
- Some routed prompts reportedly contained Chinese police systems, national ID numbers, corporate data, and surveillance information, raising national security concerns.
- DeepSeek is simultaneously scheduled to brief the UN Security Council on AI risks, adding diplomatic weight to the moment.
Why it matters
- Chinese AI stocks sold off hard: Zhipu AI fell 12%, MiniMax 6.8%, Alibaba nearly 5%, Xiaomi 3.8%, and Tencent 2.9%, signaling broad investor fear of a regulatory crackdown across the sector.
- The probe exposes a direct contradiction: Chinese AI firms used a U.S. rival's model to train their own, undermining Beijing's self-sufficiency narrative and potentially violating China's data security law.
- U.S. Treasury Secretary Scott Bessent warned that sanctions could follow if distillation is confirmed, turning a corporate IP dispute into a potential geopolitical flashpoint.
- The investigation runs parallel to a SASAC audit of Broadcom networking switches in state-owned data centers, where penetration is estimated at 90%, signaling Beijing is simultaneously tightening control over both AI software and hardware supply chains.
- Regulatory uncertainty is already slowing AI model development sentiment in China, per Gavekal Capital, as the scope of enforcement remains undefined.
What to watch next
- Whether CAC confirms violations and announces enforcement actions against DeepSeek or Moonshot, which would set a precedent for how China polices domestic AI firms using foreign model APIs.
- The U.S. Treasury sanctions decision: Bessent said a determination was coming in days to weeks, and confirmation of distillation could trigger restrictions that reshape how Chinese AI labs access Western models.
- SASAC's next move on Broadcom switches: informal guidance to remove them from state data centers would hand billions in revenue to domestic players like H3C and Ruijie, and signal how aggressively Beijing is accelerating supply-chain nationalization.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.