A new report published by the AI research firm Anthropic alleges that multiple China-based artificial intelligence companies have been engaging in 'distillation' campaigns. Distillation is a process where a smaller, more efficient AI model is trained to mimic the outputs and behaviors of a larger, more sophisticated model, effectively allowing the smaller model to 'learn' the capabilities of the original without having access to its underlying architecture.

According to Anthropic, these activities have been carried out by Alibaba, Moonshot AI, and DeepSeek. The report suggests that these efforts have intensified in recent months as the global race to develop advanced AI models has become more competitive. Anthropic characterizes these actions as a persistent effort to gain unauthorized access to the proprietary knowledge embedded within its models.

While Anthropic has publicly detailed these allegations, the companies named—Alibaba, Moonshot AI, and DeepSeek—have not yet provided a comprehensive public response or counter-statement to the specific claims made in the report. The technical nature of distillation makes it difficult to definitively prove intent, as the process can sometimes overlap with standard model training techniques. Anthropic’s report serves as a formal accusation of intellectual property misuse, highlighting the growing tensions regarding data security and model integrity in the international AI market.