Chinese AI Startups Are Mining Claude For Data.

Anthropic has leveled serious allegations against three prominent Chinese AI startups—DeepSeek, MiniMax, and Moonshot AI—accusing them of orchestrating a large-scale data extraction campaign from its Claude AI assistant. According to the company, these firms created approximately 24,000 fraudulent accounts to bypass usage policies and prompted Claude a staggering 16 million times. The targeted technique, known as model distillation, was specifically focused on mining Claude's most sophisticated capabilities, particularly its coding and complex reasoning outputs. This incident highlights the intense, global competition for advanced AI capabilities and the emerging gray areas around using one model's outputs to train another, potentially rival system.

Anthropic's complaint suggests the activity was a deliberate effort to reverse-engineer or enhance the Chinese firms' own competing models, which are key players in China's push for AI sovereignty. The scale of the alleged operation—24,000 accounts and millions of prompts—points to a systematic, industrial-level effort rather than isolated research. This case brings to the forefront critical questions about intellectual property in the age of generative AI, the ethics of data sourcing for model training, and the potential need for new technical or legal safeguards. The outcome could influence how AI companies protect their models and shape the competitive dynamics between Western and Chinese AI developers moving forward.