Anthropic says three of China's biggest AI labs have been quietly copying Claude's brains.
In a report published Thursday, Anthropic accused Alibaba, Moonshot AI, and DeepSeek of running distillation campaigns against its models. Distillation means querying a rival's AI at scale and using its answers to train a cheaper model that mimics the original's behavior - essentially letting someone else do the expensive work of building a frontier system. Anthropic says these are not isolated incidents but a sustained pattern of extraction. The company adds that the activity has escalated in recent months as competition among AI labs has intensified.
For Anthropic, this is a business problem dressed up as a security one. If a rival can approximate Claude's outputs on the cheap, it erodes the return on the enormous compute and research spend that went into training it, without any licensing fee changing hands. Calling the practice out publicly also puts distillation on the table as a competitive and policy issue, not just a technical nuisance to patch quietly.
Naming names is the notable part here - vague warnings about bad actors are common in this industry, but pinning specific companies to specific behavior invites scrutiny of Anthropic's own incentives as much as the accused.