China's free AI model is tearing Trump's tech advisors apart

Key points

• Chinese AI company Moonshot launched Kimi, a free, open-source AI model, in the week of 14 July 2025, and it appears to match the intelligence of paid models from OpenAI and Anthropic.
• David Sacks, who served as the White House AI and crypto advisor until March 2025, publicly called Anthropic's models "lobotomized" and "woke" in response to the debate.
• Emil Michael, a senior Pentagon official, called OpenAI's new head of strategic futures a "supreme village idiot" in the same online dispute.
• The Trump administration announced measures in April 2025 to curb distillation, a process where AI models are trained on the outputs of rival AI models, which US companies say Chinese firms use to copy their work.
• New York became the first US state to ban new data centres in the week before these events, adding to a wider mood of distrust toward big AI companies.

A free chatbot from China has done something remarkable: it has turned the people who advise Donald Trump on artificial intelligence against each other, publicly and messily.

The model is called Kimi. Chinese company Moonshot launched it last week as a free, open-source tool, meaning anyone can use or build on it at no cost. Early tests suggest it is roughly as capable as the paid models sold by OpenAI and Anthropic, two American AI companies whose products cost money to access. That gap matters a great deal.

Every time a powerful free model arrives from China, businesses have less reason to pay for the American alternatives. Those American companies have been propping up a big slice of US economic growth, so a credible free rival from Beijing is both a business problem and a political one for the White House.

So what is the Trump administration actually going to do about it?

Honestly, nobody agrees. That is the whole problem.

On one side sits David Sacks, who ran AI and crypto policy at the White House until March this year. Sacks argues that Chinese models have become popular partly because they have fewer restrictions on how people can use them, and he has criticised leading US AI companies for wanting the government to shut out open-source, or freely available, competition on their behalf.

On the other side, more hawkish voices inside the administration think AI models are now powerful enough to pose genuine national security risks, and that the government must control what gets released and when. That view has produced a new White House review process designed to vet AI models before they go public. Dean Ball, a former Trump AI advisor who now works for OpenAI, called that process a "de facto licensing regime" and suggested Trump might simply use quiet pressure to discourage US companies from touching Chinese models.

That suggestion did not land well. Emil Michael, a senior Pentagon official, publicly called Ball a "supreme village idiot" and insisted the government would act through democratic channels rather than behind-closed-doors pressure.

Meanwhile, MIT Technology Review points out what the arguing has drowned out: nobody is clearly explaining how Kimi got so good in the first place. China has limited access to the specialised chips that train AI models, thanks to US export controls, though those controls loosened recently when the Trump administration allowed chip-maker Nvidia to sell more hardware to China. There are also allegations of chip smuggling. One possibility is that Moonshot used distillation, a technique where a new AI is trained on the answers produced by an existing, more powerful AI, essentially learning by copying. OpenAI and Anthropic have long complained about this practice, and the Trump administration announced steps against it in April 2025.

None of that stopped Kimi from arriving. It is out, it is free, and it is nearly as capable as an Anthropic model the US government once considered so powerful it briefly restricted access on national-security grounds.

The weekend's public row suggests people close to Trump recognise something important has happened. They just cannot agree on what to do next.