The US Just Declared War on Chinese AI Models — Not on Chips, on the Software

The US Treasury Secretary just announced that Chinese AI models might be next on the sanctions list. Not the chips. Not the hardware. The models themselves.

Treasury Secretary Scott Bessent said on Fox Business Tuesday that the US would examine Chinese open-source AI models for signs of intellectual property theft, threatening sanctions against Chinese AI companies if IP theft is established. The comments, first reported by Bloomberg, mark a significant escalation in the technological competition between the United States and China — one that moves beyond hardware controls into the software layer itself.

"We've seen a lot of talk about open source models coming and threatening the large language models in the US," Bessent said. "This administration supports open source models, but what we do not support is IP theft. If we see, especially, that overseas models are stealing from our great companies, we have the ability to sanction them because of this theft."

The Shift From Hardware to Software

For the past several years, the US strategy to slow China's AI advances has focused on hardware: restricting access to advanced chips like Nvidia's H100 and A100, tightening export controls on semiconductor manufacturing equipment, and leaning on allies like the Netherlands and Japan to close loopholes. The assumption was that without cutting-edge hardware, Chinese AI models would fall behind.

That assumption has been tested. Chinese AI labs — most notably Moonshot AI with its Kimi K3 model, released July 16 as the largest open-weight model yet — are producing systems that credibly compete with frontier US models. Kimi K3 arrived with capabilities that some observers, including OpenAI's head of strategic futures Dean W. Ball, could not dismiss as mere distillation of US models.

The hardware bottleneck is being bypassed through efficiency, scale, and a different approach to model architecture. And Washington is now signaling that if it cannot stop Chinese AI through chips, it will try through sanctions on the software itself.

The Distillation Debate

At the heart of the sanctions threat lies a contentious technical question: does model distillation constitute theft?

Model distillation is a technique that transfers some of a larger model's capabilities into a smaller, more efficient system. It is widely practiced across the industry — including by US companies — and there is no consensus on whether it constitutes intellectual property theft.

Microsoft CEO Satya Nadella recently weighed in on the debate, criticizing large AI labs for what he sees as hypocrisy. "While the great innovation that comes from model providers having fair use rights to train models on public data is needed," Nadella said, "I find it ironic that the status quo is to then turn around and impose restrictive terms on distillation."

The legal landscape is further complicated by the fact that US AI labs themselves face significant copyright challenges. Anthropic this week received court approval for its $1.5 billion settlement with authors who accused the company of illegally downloading and storing millions of copyrighted books to train its AI. The irony is unavoidable: US companies are simultaneously being sued for training on copyrighted data and threatening to sanction Chinese companies for doing something similar.

The Administration's Internal Battle

The sanctions threat is not a unified position within the Trump administration. Axios reported on Monday that the administration is considering a wholesale ban on Chinese open-source AI models, though other officials have disputed that claim.

The divide reflects a deeper philosophical split. On one side are those who see open-weight AI as a national security threat that must be contained through aggressive measures — sanctions, export controls, and procurement restrictions. On the other are advocates of open-source development who argue that restricting Chinese models would fragment the global AI ecosystem and harm US interests by creating a bifurcated internet where US and Chinese AI systems operate in separate spheres.

As artificialintelligence-news.com reported, enterprises evaluating Chinese open-weight models face a question that has nothing to do with benchmarks: whether using one will still be straightforward in a year. The uncertainty alone may be enough to push regulated enterprises away from Chinese models, regardless of what formal policy emerges.

What Comes Next

The sanctions threat, if enacted, would be unprecedented. While the US has restricted Chinese access to AI hardware through export controls, it has never directly sanctioned AI models themselves as vectors of IP theft. The move would raise complex legal questions about how to determine whether a model was "stolen," how to enforce sanctions against software that can be copied infinitely, and whether the distinction between open-weight and proprietary models holds up under geopolitical pressure.

Some in the industry argue that the focus on Chinese models is misplaced. Dean W. Ball, who served as a senior AI adviser in the Trump White House before joining OpenAI, predicted that the administration's best strategy would be to create regulatory risk around Chinese open-weight models — not through formal bans, but through soft guidance suggesting such models may contain backdoors. "It needn't be that well justified," he wrote. Enough uncertainty, and regulated enterprises retreat on their own.

David Sacks, co-chair of the President's Council of Advisors on Science and Technology, pushed back sharply, arguing that weaponizing regulatory uncertainty as a competitive tool should be unacceptable. He added that the leading closed labs, already a duopoly in model revenue, want the government to remove their open-source competition.

The bottom line: the US-China AI competition has entered a new phase. The era of fighting over chips is giving way to a fight over the models themselves. And the outcome will determine not just which country leads in AI, but whether the concept of open-weight AI survives the geopolitical pressures of a divided world.

Sources

• TechCrunch: US threatens sanctions against Chinese AI models over IP theft
• Artificial Intelligence News: Chinese open-weight models are cheap. Washington is deciding what that costs.
• Axios: The secret Trump administration battle to fight Chinese AI
• ArsTechnica: Anthropic's $1.5B copyright settlement approved