Everyone knows Chinese AI is catching up fast. The question is how.
In the last few weeks, a quiet technique called "distillation" has become the loudest argument in the AI world. A Chinese lab called Moonshot AI released a model named Kimi K3 that users quickly found to rival the top-tier commercial offerings from Anthropic and OpenAI. US officials say Moonshot got there by stealing.
Distillation is a way to train a smaller, cheaper AI model by learning from a bigger, more capable one. Google's head of AI, Jeff Dean, called it a "key technique for making smaller models more capable" that requires the frontier model as a starting point.
The technique is not new. Shashi Bellamkonda, a research director at Info-Tech Research Group, said distillation is "practiced all the time" to improve models efficiently. Many companies have used it, including Nvidia, which incorporated distillation into the development of its Llama Nemotron family of models.
Get the market news that matters in a five-minute read with Market Briefs, our free daily newsletter
But the White House sees it differently. Some government officials believe distillation explains how Moonshot was able to advance so rapidly.
The pushback came fast. More than 20 tech companies including Nvidia, Microsoft, Meta, and Palantir signed a letter urging policymakers to avoid "premature restrictions" on open-weight AI models that would "stifle competition or drive innovation overseas."
Anthropic, the company behind the allegedly stolen model, has a different view. In February, Anthropic stated that Chinese firms DeepSeek, Moonshot, and MiniMax had used roughly 24,000 fake accounts that produced 16 million interactions, effectively distilling the company's capabilities on an "industrial scale." Anthropic described halting unauthorized distillation as a matter of national security, noting that it along with other U.S. AI firms puts in place protections to prevent governments and other entities from employing AI for biological weapons development or harmful cyber activities.
Both OpenAI and Anthropic have banned distillation in their terms of service. SecurityPal founder Pukar Hamal compared distillation to copying someone else's homework, saying "It's almost like someone went to the lectures, read the textbook, and did all the hard work of doing the homework. Then some other student is like, 'Hey, I didn't do that. Can I just copy your work?'"
Box CEO Aaron Levie, one of the signatories of Friday's letter, remarked in an interview that for U.S. companies to remain competitive, they must have access to the best technology regardless of its origin. "Generally the arc is going to be that the more innovation that there is, whether that's from the U.S. or China or otherwise, you should expect more AI progress, and generally it'll bend toward being even lower cost and more efficient over time," Levie said.
This dispute is not just about one technique; it reflects a fundamental divide in how AI development is governed. Open-weight models - those whose parameters are publicly available for download and modification - have been a driving force behind rapid advances in research and startup innovation. Companies like Meta have released such models openly, while frontier labs like OpenAI and Anthropic keep their most capable systems closed.
Chinese labs, lacking access to the same computing resources and proprietary training data, have increasingly turned to distillation as a shortcut to match US capabilities. The White House views this as intellectual property theft, while many tech firms argue that restricting open-weight models would only push AI development into less transparent environments.
The conflict underscores a broader struggle between fostering open AI development and protecting intellectual property. US companies like OpenAI and Anthropic argue that distillation undermines their investments in safety and innovation, while others in the industry fear that overly strict rules could cede leadership to China.
Join Market Briefs, our free daily newsletter, for a quick daily rundown of the markets
