AIster · The agentic age · Culture
Anthropic accuses three Chinese labs of distilling Claude
Copying a rival model through its own answers became an open fight between labs
Key facts
- Date
- 23 Feb 2026
- Era
- The agentic age
- Interest
- Culture
- Threads
- Safety & governance, Labs & people
- Accounts
- About 24,000 fraudulent accounts
- Exchanges
- Over 16 million in total
- By lab
- MiniMax 13M+, Moonshot 3.4M+, DeepSeek 150,000+
Anthropic said DeepSeek, Moonshot AI and MiniMax had run “distillation attacks” on Claude, training their own models on its outputs. It said the labs used about 24,000 fraudulent accounts to generate more than 16 million exchanges. Anthropic said it was building classifiers (automated systems that flag this kind of use) to detect such campaigns and sharing indicators with other companies and authorities.
Sources
- Anthropic, Detecting and preventing distillation attacks, 2026-02-23