A new Chinese AI model has turned a policy fight in President Donald Trump’s orbit into open combat, with current and former advisers attacking major US AI companies and each other in public.
The trigger is Kimi, a free, open source model released last week by Chinese AI company Moonshot. The model appears to perform near the level of systems from OpenAI and Anthropic, whose frontier models are paid products. That creates an awkward problem for an administration tied to the fortunes of US AI firms: if free Chinese models keep getting better, American customers have less reason to pay for US alternatives.
Anton Leicht, a fellow at the Carnegie Endowment, wrote on X that Chinese AI models are “a threat for an administration that really doesn’t want more economic bad news.” The Wall Street Journal has reported that Chinese AI releases have already unsettled US stocks.
A fight over protection and permission
David Sacks, who served as Trump’s AI and crypto czar until March, attacked Anthropic on X, calling its models “lobotomized” and “woke.” Sacks also criticized major AI companies that, in his words, “want the government to eliminate their open source competition.”
Sacks has argued that Chinese models are gaining users partly because they impose fewer limits on what people can do with them, while setting aside the issue of Chinese state censorship built into those systems. His position favors more open AI development and less protection for incumbent US model makers.
That view no longer appears to dominate inside the administration. Trump’s White House has moved toward a larger government role in model oversight, driven by the argument that advanced AI systems may create national security risks. CNBC reported that the White House has created a review process intended to check AI models for security issues before release.
Dean Ball, a former Trump AI adviser who now works for OpenAI, criticized that review process on X as a “de facto licensing regime for frontier AI.” Ball also suggested Trump could address Chinese open source models through soft power, including by making US companies reluctant to use systems such as Kimi.
Emil Michael, a senior Pentagon official and a key liaison between the Defense Department and AI companies alongside Defense Secretary Pete Hegseth, rejected that suggestion. On X, Michael called Ball the AI industry’s “supreme village idiot” and said the government would use “the democratic process not some Deep State scheme.”
Chips, distillation and the Kimi question
The policy fight is sharpened by an unresolved technical question: how Moonshot built Kimi at its reported level of capability while China faces limits on access to high-end AI chips.
During much of the Biden administration and early in Trump’s second administration, US policy prioritized restricting China’s access to top AI processors. Those controls have since loosened. Trump allowed Nvidia to sell more chips to China in exchange for the US government receiving a cut, a decision that drew controversy. The Justice Department has also alleged that some AI technology was exported illegally through smuggling.
The chips used to train Kimi have not been established. One possible method is distillation, where a model is trained on the outputs of other AI models rather than only on raw data. OpenAI and Anthropic have accused Chinese AI companies of using that technique and asked the US government to intervene. In April, the Trump administration announced measures aimed at curbing Chinese firms’ use of US AI model outputs, NPR reported.
Kimi is already public and free. The dispute now is less about whether Chinese open models will arrive, since they have, and more about whether Trump’s team responds by protecting US firms, restricting releases, discouraging use of Chinese systems, or accepting tougher open source competition.
This story draws on original reporting from MIT Technology Review.
