Anthropic accuses Chinese AI labs of Claude data extraction
4.7 Security Anthropic claims three Chinese AI labs used 24,000 fake accounts to extract 16 million Claude AI responses, fueling debate over data use and IP rights.
Anthropic has accused three Chinese AI labs—DeepSeek, Moonshot, and MiniMax—of orchestrating large-scale campaigns to extract data from its Claude AI model. According to Anthropic, these labs created over 24,000 fake accounts and generated more than 16 million interactions with Claude, using a technique called distillation to harvest outputs for training their own AI systems. The company claims this activity violated its terms of service, bypassed regional restrictions, and undermined safety safeguards. MiniMax reportedly led with 13 million exchanges, Moonshot with 3.4 million, and DeepSeek with 150,000, the latter using sophisticated prompts to elicit detailed reasoning from Claude. Anthropic’s revelations have sparked debate about intellectual property and fair use in AI. Critics note the irony of Anthropic’s complaint, given the industry’s reliance on publicly available internet data for training. The incident highlights growing tensions over protecting proprietary AI technologies amid global competition.
