Anthropic accuses Chinese AI firms of data copying using fake accounts and AI distillation methods

U.S. AI company Anthropic has accused three Chinese AI firms of secretly taking data from its Claude AI system. The three companies named are DeepSeek, Moonshot AI, and MiniMax. Anthropic said these companies created more than 24,000 fake accounts on its platform to access Claude. Using those accounts, they sent over 16 million prompts to Claude to collect responses and learn from them.

Anthropic claims this was done to train and improve their own AI models faster and cheaper. The activity is called “distillation,” which means copying knowledge from a powerful AI to build another model quickly, as stated by The Wall Street Journal. Anthropic said distillation itself is legal and useful when companies use it for their own systems, but it can be misused to copy competitors.

According to Anthropic, the scale of activity differed:

• DeepSeek: about 150,000 interactions
• Moonshot AI: over 3.4 million interactions
• MiniMax: around 13 million interactions

Representatives of the three Chinese companies did not respond to requests for comment. Earlier this month, OpenAI also accused DeepSeek of using similar data-copying methods in a memo sent to U.S. lawmakers. Many Chinese AI firms, including Moonshot and MiniMax, recently launched advanced AI models with strong coding and reasoning abilities.

China AI growth concerns

DeepSeek is also preparing to release its next-generation AI model soon. When DeepSeek became popular last year, experts worried China could catch up quickly with U.S. AI companies even without top-level chips. In a past research paper, DeepSeek said it trained its model using webpages and ebooks, but some of those pages contained AI-generated answers, as per the report by The Wall Street Journal. This means its model may have indirectly learned from other powerful AI systems through internet data.

Synthetic data, including distillation methods, is becoming common because AI companies are running out of high-quality training data. Moonshot itself confirmed in a technical report that it used synthetic data to train one of its AI models. Anthropic warned this issue could become a national security risk for the U.S. The company said copied AI technology could potentially be used in military, intelligence, or surveillance systems. Overall, the case shows that the global AI race between U.S. and Chinese companies is becoming more intense.

FAQs

Q1. Why did Anthropic accuse Chinese AI companies?

Anthropic said some Chinese AI firms created fake accounts and used millions of prompts to take data from its Claude AI system.

Q2. What is AI distillation and why is it controversial?

AI distillation means copying knowledge from a powerful AI model to train another model faster and cheaper, which can raise fairness and security concerns.