Anthropic Reveals Chinese AI Labs Stealing Claude Data, for What Purpose?
US-based artificial intelligence (AI) company Anthropic has revealed large-scale unauthorised efforts by several China-based AI laboratories to train their models using outputs from Claude. The threat intelligence report, released on Thursday (10/9), names major players such as Alibaba, Moonshot, and DeepSeek.
Anthropic described this practice as ‘illicit distillation’. This process involves using the outputs from more capable AI models to train other models in order to replicate their capabilities without authorisation. Anthropic stated that it has detected and halted these activities, which took place between December 2025 and August 2026.
According to the report, the operation involving Alibaba affiliates was the largest distillation campaign ever measured by Anthropic. It recorded over 151 million data exchanges with Claude between May and July 2026. This activity peaked at nearly 3 million exchanges per day through more than 3,500 fake accounts.
Alibaba is alleged to have used Claude’s outputs to train its Qwen model, as well as for broader AI research, including reinforcement learning and model architecture.
Meanwhile, Moonshot AI—the company behind the Kimi model—was reported to be secretly forwarding Kimi user requests to Claude. Users believed they were interacting with Kimi, when in fact they were receiving responses from the Claude Opus model. Anthropic noted that Moonshot forwarded nearly 300,000 customer requests over a 10-day period through a network of 5,380 fake accounts located in Singapore and Japan.
Anthropic highlighted that many of these data exchanges contained sensitive information, including data from individual users, large multinational corporations, and even state-affiliated actors. “These practices are likely inconsistent with privacy laws and the terms of service of the laboratories themselves,” the report stated.
In addition to Alibaba and Moonshot, DeepSeek was also cited as using similar tactics. Anthropic observed more than 12 million distillation attacks linked to DeepSeek over a 14-day period in July 2026. DeepSeek transferred data exchanges to Claude without notifying its customers.
As of the time of publication, Alibaba, Moonshot, DeepSeek, and Xiaomi have not provided official responses regarding the allegations contained in the Anthropic report.