Trump's AI Advisors Turn on Each Other as China's Free Kimi K3 Exposes a Strategy With No Exit

Kimi K3 Put Trump's AI Strategy in a Trap With No Exit

Last weekend, three men who shaped the Trump administration's AI policy publicly attacked each other and the companies they were supposed to guide. David Sacks, who served as the president's AI and crypto "czar" until March, called Anthropic's models "lobotomized" and "woke." Emil Michael, a top Pentagon official who works with Secretary of Defense Pete Hegseth as the agency's liaison to AI companies, branded OpenAI's head of strategic futures Dean Ball the AI industry's "supreme village idiot." 1

The trigger was Kimi K3, a free, open-weight model from Chinese AI company Moonshot that launched last week and appears to rival models from OpenAI and Anthropic in capability. 1 2

The reason the fighting turned personal is structural. The administration's AI coalition rests on four promises that Kimi K3 just proved are mutually exclusive: champion free markets, restrict Chinese technology, protect US AI companies' revenue, and maintain national security. Pick any two and you can build a coherent position. The problem is that each faction picked a different pair, and every faction is right about why the others are wrong.

Sacks chose free markets over protectionism. On July 19, he criticized top AI companies that "want the government to eliminate their open source competition" and argued Chinese models are gaining traction partly because they carry fewer usage restrictions. 1 He is out of power, though. His deregulatory stance has been replaced inside the administration by one that sees a larger government role in controlling AI models deemed threats to national security. 1

Ball, who moved from a Trump AI advisory role to OpenAI, chose protecting US companies. He argued the government should create regulatory fear around open-weight Chinese models because they deter capital spending by frontier labs. Tech luminaries including Yann LeCun and Martin Casado argued that open software can accelerate innovation and coexist with proprietary projects, and Ball retracted those claims. 2 He separately criticized the White House's new security review process for AI models as a "de facto licensing regime for frontier AI." 1

Michael chose national security and the democratic process. He rejected Ball's suggestion that the government should quietly pressure companies into avoiding Chinese models, insisting the matter should go through "the democratic process not some Deep State scheme." 1

The economic stakes behind the infighting are concrete. Every time a capable free model like Kimi K3 arrives, companies have less reason to pay for access to OpenAI or Anthropic's models. AI industry enthusiasm is driving an outsized share of US economic growth, and Chinese models have already rattled US stocks. 1 Braden Hancock, co-founder of Snorkel AI and a former Meta Director of AI, told TechCrunch that frontier-caliber open source models will squeeze the margins and depress prices at the companies that build them. 2

So what can the administration actually do? Axios reports it is considering banning Kimi K3 and other advanced Chinese models at the behest of American frontier labs. Politico reports the Department of Commerce will not take that step anytime soon. 2

A software ban would be an extraordinary move for an administration that champions free markets and open innovation. It would restrict tools that US companies actively want to use, including for security tasks that American models refuse to perform. Sacks has highlighted cases where US companies turned to Chinese LLMs to close security gaps when domestic models declined. 2

The strategy that was supposed to prevent this moment has already frayed. The US spent years trying to keep advanced chips out of Chinese hands. Those export controls loosened when Trump allowed Nvidia to sell more chips to China in exchange for the US government taking a cut. 1 It is unclear what chips Moonshot used to train Kimi K3. The process may have involved distillation, where models are trained on the outputs of existing AI systems. OpenAI and Anthropic have long complained about Chinese companies doing this, and the Trump administration announced efforts to curb the practice in April. 1

None of it stopped Kimi from arriving. The model is free, openly available, and MIT Technology Review reports it is nearly as good as the Anthropic model the US government deemed so powerful it was briefly shut down for threatening national security. 1

The weekend's infighting is the first visible crack in a strategy that has no resolution. Ban the software and betray free-market principles while alienating the open-source community the administration claims to champion. Allow it to compete and watch the companies driving US economic growth bleed revenue and stock value. The national security argument cuts in both directions, which is exactly what the factions cannot agree on.

Sam Bresnick, a research fellow at Georgetown's Center for Security and Emerging Technologies, frames the bind plainly: he asks why the weight of the US government should be aimed at protecting companies from competitors excluded from the US market based on their origins. 2

Nobody in Trump's orbit has an answer. And last weekend, every faction proved it.