Anthropic revealed that several major Chinese AI labs ran coordinated campaigns to extract knowledge from Claude. Nearly 190 million exchanges were harvested between May and July 2026. The bulk came from operators linked to Alibaba (over 151 million), while DeepSeek contributed over 12 million exchanges and Moonshot routed more than 300,000 requests in just 10 days. The actors used thousands of fraudulent accounts to extract detailed reasoning traces from Claude, aiming to improve their own AI models. Anthropic countered by banning organizations, enforcing identity verification, and limiting the detail in reasoning outputs.
Source: Read the original article
