This US cyber defense agency report was published ahead of the US-China presidential meeting scheduled later this month at the White House, where AI could become an important topic of discussion.
This is the latest in accusations from the US government and American AI companies regarding Chinese AI thefts carried out using the so-called distillation process.
“AI companies based in China systematically extract patented features and capabilities of US AI company models,” the Tuesday report states. They do this “through industrial-scale knowledge distillation campaigns that form the core of their AI development strategy, not just its supplement,” the report says.
Distillation is a process where an AI model is trained to mimic the results of a larger and more powerful model. This method is widely used across the industry, but Washington claims that Chinese companies apply it on a large scale to illegally copy American systems.
Chinese companies employing this strategy “shorten AI development time and reduce financial costs associated with training advanced models,” the report states.
It details how “DeepSeek,” “Moonshot AI,” “Alibaba,” “MiniMax,” “StepFun,” and “Z.AI” allegedly exploited millions of queries to learn the technical capabilities of US models such as “OpenAI,” “Anthropic,” “Google,” and “xAI.”
These actions, reportedly “most likely with the knowledge of the Chinese government,” have been ongoing at least since the end of 2024.
D. Trump is under pressure to respond to the growing popularity of Chinese AI models, as many companies choose Chinese models over more powerful but more expensive US versions developed by “Anthropic” or “OpenAI.”
Read more Marius Ivaškevičius: «The Dawn of the Gods» was provoked and dictated by war»
