At the start of the Cold War, when the United States and the Soviet Union were engaged in a relentless nuclear arms race, they viewed nuclear technology as a tool of soft power and diplomacy – something to share and export, a seal of alliances and a sign of friendship.
So, in the same year that it executed the Rosenbergs for revealing our nuclear secrets to the Soviets, Dwight Eisenhower’s America launched the Atoms for Peace program, which helped establish civilian nuclear programs around the world. The Soviet Union also exported forms of nuclear energy to Eastern Bloc countries and Mao’s China. In each case, the overarching goal was to spread the technology while maintaining control of the weapons, with very different approaches taken to the two forms of nuclear power.
This story suggests a framework for understanding the difference between the American and Chinese approaches to artificial intelligence. The United States is behaving as if frontier AI models could be the equivalent of nuclear models Weaponsoffers an immensely destructive form of power that requires careful and jealous observation. American AI tycoons constantly talk like they’re engaged in an Oppenheimer-style race (complete with Oppenheimer-style moral hand-wringing), the American military-industrial complex is eager to stop China from closing the current AI gap, and both industry and government are increasingly viewing border models as serious security threats.
China now treats the models its companies produce as equivalent to nuclear weapons energya lower-risk technology to be shared and commercialized to win friends, influence the world, and undermine the advantages of America’s AI giants. There is some lip service to risk and security in China, but little sense that this is a unique power to be hoarded, a geopolitical asset, or a weapon with existential risk.
For a while (well, a few months, but that’s an eternity in AI time) I wondered whether the Chinese position was sincere. Given that its companies and models have consistently lagged behind Anthropic and OpenAI, China clearly has an incentive to use open source to undermine Silicon Valley’s profits and reach, while at the same time Washington must worry about the risks that the most advanced models could pose. If China suddenly surpasses us, I speculated, its incentives might change and it might seek to seal off and weaponize its best models, treating them as state secrets and sovereign assets.
We don’t yet have a Chinese model that beats our own, but this week brought a model that perhaps comes close, Kimi K3 from Chinese company Moonshot AI. And so far, Beijing is treating the Kimi K3 like all its predecessor models: it will be open source to ensure maximum acceptance, without worrying about risks, guard rails and export controls like the American model.
This decision should make us a little more confident that Beijing’s overall approach is sincere. It suggests that the Chinese are not just rejecting the utopian and dystopian scenarios common in Silicon Valley, where, beyond a certain threshold, the power of artificial intelligence is completely reshaping the world. They may not even be particularly “risky” in the sense that they are worried about the kinds of dangers that today’s AI models could already pose—through the ability to hack and collapse complex systems, the ability to help terrorists create bioweapons, the uncertain risks of fraudulent AI behavior, and more.
American AI companies and the national security state seem to assume that we are already at a point where you simply cannot allow the world to use the best AI models without controls and guardrails, any more than you would send nuclear bombs or ICBMs to random households or private companies. While the Chinese are willing to shrug their shoulders and push the envelope and let America worry about security and sovereignty, they assume that AI is a normal technology and the security concerns will be resolved over time.
Or to return to a more cynical reading: perhaps they are making a double bet. If AI poses little risk and is not actually some kind of strategic superpower, its open source models are the best way to undermine the American advantage and make the world more like China than us. Alternatively, if we get some sort of AI Chernobyl from someone abusing a Chinese model, the fallout and backlash will likely still hit US companies the most, as the American public is already skeptical of AI and prepared for data center moratoriums.
https://www.nytimes.com/2026/07/18/opinion/china-united-states-artificial-intelligence.html
