Anthropic has published a report accusing several Chinese artificial intelligence companies of systematically copying its technology. The company says Alibaba, Moonshot AI, and DeepSeek have run persistent distillation campaigns against its models, and that these efforts have grown more intense as global AI competition heats up. This is one of the most direct public accusations a major Western AI lab has made against Chinese rivals.
Model distillation is a technique where developers repeatedly query a powerful AI model and use the responses to train a smaller, cheaper model. While distillation is common in the industry, doing it without permission against a competitor's service is widely seen as unethical and a violation of terms of use. Anthropic says the campaigns have been going on for some time but have escalated sharply in recent months as the race to build capable AI has intensified.
The report names three of China's best-known AI labs. Alibaba is a major cloud and e-commerce giant with its own large language models. Moonshot AI and DeepSeek are also prominent players in China's fast-growing AI scene. By distilling Anthropic's models, these companies could potentially close the quality gap without investing the same massive resources in training from scratch.
This matters because the global AI race is increasingly tense. If companies can simply copy the outputs of American labs, it undermines the incentive to spend billions on original research. It also raises security concerns, since bad actors could use similar techniques to replicate safety guardrails or remove them entirely.
Anthropic is calling attention to the issue publicly, which suggests it wants industry-wide pressure or policy responses. The company may also tighten its technical defenses to make distillation harder. For now, the report puts a spotlight on a shadowy practice that could shape how AI companies protect their most valuable assets in the years ahead.