American authorities have accused six Chinese AI companies, including DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI, of using a technique known as "model distillation" to extract capabilities from advanced American AI systems and employ them in developing their own models.

According to the US authorities, the targeted models included Claude from Anthropic, GPT from OpenAI, Gemini from Google, and Grok from xAI. A large number of requests were implemented with the aim of collecting outputs from these systems and using them to train competing models.

What is the story behind "distilling" AI?

The technique of distillation relies on using the output of a powerful model to train another model. This technique is well-known in AI development and is not inherently illegal, but Washington claims that the problem lies in the widespread use of this method and the methods used, which violate the terms of use of the models or extract their capabilities without authorization.

These accusations come at a time when competition between the United States and China to develop artificial intelligence technologies is increasing. China has denied the US allegations, describing them as baseless, and warned against using this issue to restrict Chinese companies and competition in the AI market.

This case opens up a new confrontation over intellectual property rights for AI, and the extent to which the answers of advanced models can be used to develop competing models. This is at a time when companies worldwide are racing to reach more capable and cost-effective systems.