Sometimes, when a Chinese company releases a new AI model, Americans panic.
That’s what happened last week when China’s Moonshot AI debuted Kim K3, which smashed a number of notable benchmarks. By most accounts, the Kimi K3 rivals leading American models at a fraction of the cost.
This new model has reignited panic that China is closing in on the US in a race to corner the AI market, and prompted accusations that China is training its models on work already done by Anthropic, OpenAI, and Google. (Earlier this month, Business Insider’s Ali Bar highlighted the irony of that accusation.)
Tensions increasingly center around clear strategic differences between the two countries. China has adopted an open source or open weight model, while the US has remained largely closed.
Read more about Moonshot’s latest models
Debate on open versus closed model
Controversy regarding X erupted over the weekend after OpenAI executives published a lengthy response to Kimi K3.
“Given the potential risks, I am personally surprised that the Chinese state continues to allow such a good model to be open sourced,” Dean Ball, a former senior AI advisor to President Donald Trump and the new head of strategy at OpenAI, wrote about X.
He said an open-weight strategy would lead to full-blown “AI communism” and that an open model could be “slowdownist” because it would “curb capital investment in AI.”
But what really got the conversation going was this:
“At some point, the Trump administration will realize that the best strategy here is to create significant regulatory risk regarding the use of China’s promiscuous model,” he wrote.
Ball suggested that fear, uncertainty, and doubt (or “FUD” for those in the know) in the regulatory process will cause most American companies to avoid using open models.
He later clarified that this was his prediction, not a recommendation, and said he supports open source until AI becomes too dangerous, at which point it would be a “sad day.”
The response was swift and widespread.
Fabricating regulatory chaos to support American AI labs is a lot like “regulatory capture.” This is when government agencies were tasked with regulating industry design rules to support the industry, often with the advice of industry insiders themselves.
Anthropic and OpenAI argue that their models are too powerful to be open. It’s dangerous when left open, allowing anyone to use the tool for any purpose with little oversight. Closed systems give manufacturers more overall control, including security, access, and pricing. The institute warns that open weight models coming out of China are a threat to national security and its business.
David Sachs, a venture capitalist who served as President Trump’s first AI and cryptocurrency czar before becoming co-chair of the president’s Science and Technology Advisory Council in March, said the “weaponization of regulatory uncertainty” is “totally unacceptable.”
“We are at a critical inflection point in AI policy. The large, closed labs, which already have a duopoly in terms of revenue from AI models, want the government to eliminate open source competition,” he wrote on X in response to Ball’s post. That “duopoly” refers to OpenAI and Anthropic. “They put their cards on the table, and it’s time for the rest of Silicon Valley, the majority of people who still value free competition, to do the same.”
Chamath Palihapitiya, co-host of the Sacks’ All-In podcast and fellow VC, was equally blunt: “The future is open source. We need to embrace it and get on with it,” he wrote about X.
Suhail Doshi, a prominent software engineer and entrepreneur, said that US AI institutes train their products on “humanity data” and do not pay them a penny.
“Lobbying and laws seeking to ban random weight models in the name of ‘distillation’ are completely illegal,” he wrote to X. “This is a fight for the future of American innovation.”
A Citrini Research analyst known as Jukan in X disagreed with Ball and his warnings about Chinese takeovers, writing that the open source model does not automatically put companies in a position of dominance. He said that DeepSeek, for example, operates more efficiently because it operates on its own as well as an open source framework, which results in lower token costs.
“Chinese companies may lack sufficient computing power to meet all the inference demands themselves, but that doesn’t mean they can’t sell at a loss or recover their training costs,” Jukan wrote.