According to a recent report by Anthropic, they have discovered and banned distillation activities related to Claude from seven Chinese AI companies over the past few months. The companies named include Alibaba, Dark Moon, DeepSeek, Zhipu, Xiaomi, SenseTime, and MiniMax. Distillation is a common training method, but Anthropic's allegations focus on large-scale circumvention of restrictions, scraping Claude's outputs and user data. 1. Alibaba: The largest scale. Anthropic claims there were over 151 million interactions from May to July, peaking at nearly 3 million per day, primarily extracting the complete reasoning process of Opus 4.6 and 4.7 for training Qwen 3.5, 3.6, and 3.7. 2. Dark Moon: Directly using Claude to answer Kimi users. Nearly 300,000 real user requests were forwarded to Claude within 10 days, some of which included internal company codes and valid credentials; interactions exceeded 23 million from May to July. 3. DeepSeek: Also secretly forwarding user requests to Claude and specifically identifying users utilizing Claude Code, OpenCode, and other tools, selectively forwarding some requests to Opus. Over 12.1 million interactions in 14 days. 4. Zhipu: Specifically extracting and cleaning Claude's reasoning process, also using American models to train cybersecurity capabilities. Anthropic noted that Zhipu attempted to distill Fable but switched to Opus 4.6 and another American company's model after finding the defenses too strong. Over 3.4 million interactions in 17 days. 5. Xiaomi: Feeding real conversations and programming records generated by MiMo users back to Claude to generate SFT and reinforcement learning data. Over 400,000 requests in 20 days; Anthropic even suspects that the free trial period of MiMo-V2-Pro may have been used to attract more overseas developers to generate training material. 6. SenseTime: Purchasing chat records between users and Claude from third-party data vendors. These data come from agents and relay services, with some users unaware that their conversations would be saved and sold. 7. MiniMax: Accused of operating a model relay service through a company that appears unrelated, providing only Anthropic and OpenAI models, without even including MiniMax's own models. Anthropic assesses that this service may be intended to collect user interactions with American models for training. All of the above are based on Anthropic's investigations, attributions, and allegations, which have not yet been independently verified.