Chinese AI models like Kimi K3 challenge US dominance, forcing Washington policy debate on open-weight vs. proprietary model control.
Over the weekend, several current and former advisors to President Donald Trump on AI publicly lobbed insults at the country's leading AI companies. David Sacks, the president's AI and crypto "czar" until March, branded Anthropic's models as "lobotomized" and "woke." Emil Michael, a top Pentagon official, called OpenAI's new head of strategic futures a "supreme village idiot."
The disagreement began over Kimi, a free, open-source model that Chinese AI company Moonshot launched last week. The model appears to rival the intelligence of offerings from OpenAI and Anthropic—which are decidedly not free. This creates a genuine problem for Trump and has fractured his AI advisory circle into opposing factions. Each time a new capable, free model from China gets released, US companies lose leverage over customers unwilling to pay for Anthropic or OpenAI. Given that enthusiasm for these companies drives an outsized share of economic growth, Chinese AI models present both economic and political headaches for the administration. As Anton Leicht, a fellow at the Carnegie Endowment, wrote on X, they are "a threat for an administration that really doesn't want more economic bad news"—and they have already rattled US stocks.
The situation is complicated by broader distrust of AI companies. New York imposed the country's first state ban on new data centers just a week before Kimi's release, and many Americans would likely have little sympathy for OpenAI or Anthropic facing cheaper competition from abroad. On July 19, Sacks himself noted that top AI companies "want the government to eliminate their open source competition," arguing that Chinese models have gained popularity partly because they come with fewer restrictions—though they do carry built-in state censorship.
However, Sacks no longer holds an official advisory role with Trump. His advocacy for freer AI has given way to an administration position favoring greater government control, rooted in the belief that AI models strong enough to threaten national security must be regulated. This shift has led to a new White House review process designed to vet AI models for security before release. Dean Ball, a former Trump AI advisor now at OpenAI, criticized this over the weekend as a "de facto licensing regime for frontier AI," suggesting Trump might use soft power—making US companies wary of models like Kimi—rather than formal restrictions. Emil Michael bristled at the characterization, insisting the government would operate through "the democratic process not some Deep State scheme" rather than quietly coercing companies.
What has been overlooked in this debate is how Kimi achieved its capabilities in the first place. Throughout the Biden administration and into Trump's second term, keeping advanced chips out of China's hands was a policy priority. Yet those export controls have loosened—Trump controversially approved Nvidia selling more chips to China in exchange for government revenue—and alleged chip smuggling has occurred. China still has limited computing power, and it remains unclear what chips Moonshot used to train Kimi.
The model may have been developed through distillation, a technique where AI models train on the outputs of existing models. OpenAI and Anthropic have long complained about this practice by Chinese competitors and sought government intervention to stop it. The Trump administration obliged in April with measures to curtail it. But Kimi exists now, available for free, and performs nearly as well as an Anthropic model the US government once deemed so powerful it briefly shut it down on national security grounds. The weekend's quarrel among Trump's advisors suggests many view this as a wake-up call—though nobody can agree on what it should prompt.