China rejects Anthropic's AI extraction allegations

China has firmly rejected accusations by US artificial intelligence company Anthropic that Chinese AI labs, including DeepSeek, Alibaba and Moonshot AI, carried out large-scale efforts to extract responses from its Claude models and use them to train their own systems.

Chinese foreign ministry spokesperson Mao Ning said Beijing “advocates AI for good” and accused Anthropic of “distorting facts”. “We firmly oppose attempts to throw mud at China by distorting facts,” Mao said at a regular news conference on Friday. The comments came a day after Anthropic released a threat intelligence report alleging that several China-based AI companies had conducted what it called “illicit distillation” campaigns against Claude.

Distillation is a common AI training technique in which a less powerful model learns from the outputs of a more capable model. Anthropic alleged that Chinese AI labs went further by using networks of accounts to systematically send prompts to Claude, collect its responses and use those exchanges to improve their own models. According to the report, the operations involved more than 151 million exchanges with Claude by Alibaba alone between May and July, making it the largest distillation campaign Anthropic said it had measured. The activity peaked at nearly 3 million exchanges a day and involved more than 3,500 accounts that Anthropic described as fraudulent. The company said Alibaba used Claude’s outputs to train its Qwen models and also used the model for wider AI research, including reinforcement learning and model architecture work. Anthropic said some of the exchanges involved sensitive information belonging to individual users, multinational companies and state-affiliated organisations. It said such practices were likely inconsistent with privacy laws and the companies’ own terms of service.

Anthropic also accused Beijing-based Moonshot AI of secretly routing Kimi users’ requests to Claude and using some exchanges to train its models. The company said Moonshot relayed nearly 300,000 requests through 5,380 accounts in 10 days and more than 23 million exchanges between May and July. DeepSeek was accused of similar tactics, with Anthropic reporting more than 12 million distillation attacks over 14 days in July. Xiaomi was also named in the report. Anthropic said some unauthorised exchanges exposed sensitive data, including internal CCTV footage from Chengdu. It said it detected and disrupted the alleged operations between December 2025 and August 2026 across seven areas of AI misuse, including cyber operations, surveillance, fraud, weapons development and model distillation.