Trump team crafts quiet kill switch for Chinese AI models

The Trump administration isn’t drafting a splashy executive order to ban Chinese AI outright. Instead, Axios reports that the Commerce Department, NSA, and White House are weaving together a quieter, more durable net of sanctions, procurement rules, and public pressure campaigns designed to make hosting models like DeepSeek or Kimi K3 too legally toxic for American companies to touch.

The strategy has been in the works since 2025. Commerce reportedly drafted supply chain rules targeting Chinese open-source models as early as last summer, though lighter-regulation advisers initially blocked them. That changed after China released the Kimi K3 model and key personnel shifted within the White House, giving sanctions hawks new leverage. One government source described the approach as “slower and more durable” than a direct ban—less splashy than an executive order, but harder to reverse.

OpenAI strategist Dean W. Ball has a name for this: FUD regulation. Fear, uncertainty, and doubt. Ball argued that just enough regulatory risk will cause “every regulated enterprise” to back away from Chinese models without needing binding rules. “You probably don’t want to create so much regulatory risk that you scare off the hyperscalers from serving Chinese models; this will just drive startups to sketchier providers,” he wrote. “There’s a happy middle ground here.” The goal isn’t prohibition—it’s making compliance so ambiguous that corporate legal departments kill the deals themselves.

The commercial stakes are enormous and barely concealed. Chinese open-source models are cheaper and increasingly competitive, threatening the market dominance of Google, OpenAI, and Anthropic. With AI stocks driving much of the market’s gains under Trump, a sudden erosion of American AI revenue wouldn’t just bruise Silicon Valley egos—it could rattle the broader economy. Security concerns about backdoors are real, but the government’s own logic contains an awkward tension. Hugging Face has shown that open models can outperform commercial ones on cyber defense tasks, meaning restrictions might actually degrade security in some contexts. The quiet part, said almost aloud, is that this is as much about protecting market share as it is about protecting networks.