Anthropic has come out with a new report claiming three major Chinese AI developers conducted huge campaigns to copy its technology. The labs, MiniMax, DeepSeek, and Moonshot, allegedly used thousands of fraudulent accounts to illicitly extract capabilities from Anthropic"s flagship model, Claude. The operation specifically targeted Claude"s most advanced skills in agentic reasoning, tool use, and coding, according to the lab.
The thing is, distillation, the technique where a weaker model learns by mimicking the outputs of a more advanced one, is a legitimate industry practice. Frontier AI labs regularly distill their own large models to create smaller, cheaper, and more efficient versions for specific customer applications. The problem, Anthropic says, is when a competitor uses the technique to acquire powerful capabilities in a fraction of the time and at a fraction of the cost it would take to develop them from scratch.
Anthropic"s report details an "industrial scale" campaign involving approximately 24,000 fraudulent accounts that generated over 16 million exchanges with Claude. This was done with the help of services that run so-called "hydra cluster" architectures, sprawling networks of accounts that mask and distribute traffic to evade detection.
The American lab claims it was able to identify the culprits with "high confidence through IP address correlation, request metadata, infrastructure indicators, and in some cases corroboration from industry partners". Here"s the full breakdown:
| Lab | Exchanges | Targeted Capabilities |
|---|---|---|
| MiniMax | > 13 million | Agentic coding, tool use |
| Moonshot | > 3.4 million | Agentic reasoning, coding, computer vision |
| DeepSeek | > 150,000 | Reasoning, rubric-based grading, censorship-safe responses |
In response to these findings, Anthropic said it has built several classifiers and behavioral fingerprinting systems designed to identify these distillation attack patterns in its API traffic. The company is also sharing technical indicators with other AI labs and cloud providers and has strengthened verification for educational and research accounts. This includes developing tools that can detect coordinated activity across large numbers of accounts.
The AI lab recently closed a massive $30 billion Series G funding round, pushing its valuation to $380 billion. It also recently released upgraded models, Claude Sonnet 4.6 and Claude Opus 4.6, boasting enhanced performance.