Anthropic Reveals! Don’t Trust Chinese AI Companies!
Anthropic explained in its report how Chinese AI companies collect user information without permission through Claude and the distillation formulas.
As competition intensifies in the world of artificial intelligence, Anthropic, a leading name in the field, has released today a comprehensive report revealing striking details about the methods used by Chinese tech giants to train their AI models. According to the report, Chinese companies, including prominent names like DeepSeek, Xiaomi, and Moonshot AI, have been found to be using a “distillation” method to train their own systems by taking questions and data that users provide to advanced models like Claude.
During this process, users’ personal data and sensitive company information are unknowingly transferred to the training pool of third-party models. Anthropic argues that this situation violates both the laws of confidentiality and contradicts the companies’ own terms of use.
- Chinese AI companies are using user data to train their own systems by directing it to models like Claude.
- Alibaba is recorded as the company that carried out the largest distillation attack in history with more than 151 million processes.
- It has been determined that companies like Moonshot AI collect data by processing customer requests directly through Claude.
- Anthropic emphasizes that this situation poses significant risks in terms of user confidentiality and ethical rules.
Chinese Companies Use Claude Data in Model Training
One of the most striking parts of the report reveals the scale of the “distillation” activities. Anthropic states that Chinese developers are taking user questions and sending them to Claude, and Claude is then incorporating Claude’s answers into their own models’ knowledge base. During this data transfer, names, email addresses, and confidential company information are being processed without protection. This poses a significant threat to data security, particularly for corporate users.
Alibaba Processes Record Levels of Data
According to Anthropic’s analysis, Alibaba alone is managing the largest data extraction attack ever seen in the AI world, with over 151 million processes. The processing of such a large volume of information demonstrates the Competitors’ efforts to transfer the reasoning capabilities of advanced models like Claude to their own local models. The company emphasizes that this activity is not only a technical competition but also a process that pushes ethical boundaries.
Moonshot AI Directs Customer Requests to Claude
It has been determined that Moonshot AI, a company known for its various model families, directs customer requests to Claude instead of resolving them directly within its own infrastructure. While this strategy is a procedure used to make systems appear smarter, the transfer of users’ information to another platform without their consent invalidates confidentiality agreements. Anthropic states that hundreds of users’ data is indirectly shared through this method, and that this is unusual for global data privacy standards.
How ethical do you think it is for AI companies to use user information in this way when training their models? You can share your opinions on these strategies of technology giants in the comments section.
