Anthropic Reveals Chinese AI Labs Attempting to Distill Claude's Features

Anthropic revealed that seven China-based artificial intelligence research institutes were attempting to use Claude's functionality for developing their own models. The company stated these institutes utilized intermediary platforms and numerous accounts to send large volumes of requests to Claude in an effort to distill the model.
Anthropic explained that Alibaba-related activities saw over 3,500 fake accounts generating more than 151 million exchanges from May to July 2026. The daily peak reached approximately 3 million exchanges, with Anthropic alleging these were aimed at incorporating Claude's features into the development of Alibaba's Qwen model.
The company also claimed that Moonshot AI conducted over 23 million exchanges, while DeepSeek transmitted sensitive customer information to Claude, according to their report. Distillation involves using outputs from a stronger model to train a relatively smaller one, and Anthropic stated this activity went beyond standard model training practices.
Korean Source
This article is an English localization of a Korean-language crypto news report. Original headline: 앤트로픽, 중국 AI 7곳 증류 정황·수법 공개