China’s Free Kimi Model Is Exposing a Trump AI Civil War

A free, open-source AI model from China has accomplished what few things can in Washington: it’s got President Trump’s own AI advisors publicly tearing each other apart. The catalyst is Kimi, a new model from Chinese firm Moonshot that launched last week and appears to rival paid offerings from OpenAI and Anthropic. Its release immediately resurfaced a bitter factional dispute over how to handle China’s accelerating AI sector.

The personal attacks were swift and vicious. David Sacks, Trump’s former AI and crypto ‘czar’ who left his post in March, branded Anthropic’s models ‘lobotomized’ and ‘woke.’ Not to be outdone, Pentagon official Emil Michael called a former Trump AI advisor now working for OpenAI a ‘supreme village idiot.’ The vitriol stems from a fundamental disagreement: Sacks’ camp favors a hands-off approach where open-source competition thrives, arguing that U.S. companies want the government to simply ‘eliminate their open source competition.’ The prevailing view now inside the administration, however, leans toward aggressive government intervention to control AI on national security grounds.

This internal war leaves the White House in a bind. Every time a capable, free model like Kimi drops, it erodes the business case for paying U.S. companies, rattling stocks and threatening an economic boom driven largely by AI enthusiasm. The timing is politically awful, coming just as New York imposed the country’s first state ban on new data centers. Anton Leicht, a fellow at the Carnegie Endowment, summed it up bluntly on X, calling the situation ‘a threat for an administration that really doesn’t want more economic bad news.’

The core technical mystery also fuels the paranoia. Despite U.S. chip export controls—which Trump has controversially loosened in a quid-pro-quo with Nvidia—China operates with limited computing power. It’s not clear what chips Moonshot used to train Kimi, feeding suspicions of ‘distillation,’ a technique where models train on outputs from existing AIs. OpenAI and Anthropic have long cried foul over the practice, and in April the Trump administration launched efforts to curb it. Yet here we are, with a model nearly as powerful as one the U.S. government briefly shut down for threatening national security. The sparring suggests a wake-up call is blaring, but nobody can agree on what they’re supposed to be waking up to.