Chinese AI labs secretly used millions of Claude exchanges to train their models, Anthropic says
Anthropic said it detected unauthorized efforts by China-based AI labs including Alibaba and Moonshot AI to use its Claude models to help improve their own AI systems.
TL;DR
- Anthropic identified unauthorized "illicit distillation" campaigns by Chinese AI labs.
- Alibaba, Moonshot, and DeepSeek allegedly used Claude model outputs to train their own AI systems.
- Alibaba's operation involved over 151 million exchanges with Claude between May and July.
- Moonshot forwarded customer requests to Claude and used the responses for training, involving nearly 300,000 requests in a 10-day period.
- DeepSeek also used similar tactics, with over 12 million distillation attacks observed in July 2026.
- These practices may be inconsistent with privacy laws and terms of service.
- Anthropic disrupted these activities across seven areas, including cyber operations and influence operations.