Surveillance footage of People's Liberation Army facilities and Chinese police records are among the sensitive files allegedly leaked to a U.S. artificial intelligence firm by domestic startups. The Cyberspace Administration of China is investigating whether DeepSeek and Moonshot AI secretly routed user data to Anthropic’s Claude models to improve their own AI systems without customer knowledge. Anthropic reported that Moonshot forwarded over 23 million exchanges to Claude between May and July, and DeepSeek routed 12.1 million exchanges in just 14 days during July. Regulators initially questioned employees from seven companies, including Alibaba, Xiaomi, and SenseTime, before focusing the probe on DeepSeek and Moonshot.

Shares of Chinese AI developers plunged following reports of the investigation. Z.AI fell as much as 12%, MiniMax dropped 6.8%, and Alibaba slid nearly 5% in Hong Kong, while broader tech stocks including Xiaomi and Tencent declined 3.8% and 2.9%, respectively. Investors are concerned that increased regulatory scrutiny on data security could slow the rollout of new models like Alibaba’s Qwen3.8-Max and Moonshot’s Kimi K3 as competition from OpenAI and Anthropic intensifies.