business 7 min read

US Plans to Force Disclosure of Chinese AI Use

The Trump administration is weighing a requirement that US companies publicly name whether they use Chinese AI models. The move creates a compliance nightmare for firms—and especially Korean chipmakers—caught between Washington and Beijing.

  • Korean Semiconductors
  • Chinese AI
  • US Policy
  • Tech Decoupling
  • Compliance Risk

The Quiet War Over Open Weights

American companies are quietly running more Chinese AI than anyone in Washington wants to admit. And now the Trump administration is considering forcing them to say so out loud.

According to Reuters, the Treasury Department and the White House Office of Science and Technology Policy have been discussing a proposal that would require US enterprises to publicly disclose on their own websites whether they rely on Chinese AI models. The strategy isn’t to ban Chinese technology outright or impose sanctions—it’s something arguably more corrosive. It’s a public naming-and-shaming regime designed to make continued reliance on Chinese AI politically costly for American corporations.

The political calculus is clear. Rather than risk accusations of being soft on China, the administration is pushing transparency as a deterrent. A public disclosure mandate accomplishes what an outright ban cannot: it externalizes the reputational risk onto individual companies, letting Washington sidestep the legal headaches and trade-war escalations that direct prohibition would trigger. Companies bear the burden themselves—figuring out what counts, how to document it, and who will notice.

Who Is Already Using Chinese AI

The list of US firms dipping into Chinese models reads like a roll call of everyday technology infrastructure. Airbnb uses Alibaba’s Qwen for customer service agents. DoorDash has integrated parts of its internal systems with Moonshot AI’s Kimi. Cursor—the popular AI coding tool—is built on Composer 2, which itself runs on Kimi. A letter from Republican lawmakers led by Andrew Garbarino and John Moolenaar specifically targeted these companies, asking them to explain their Chinese AI dependencies.

This is not a fringe problem. Data from the AI model marketplace OpenRouter shows that the three most-used AI models by token volume in October 2026 were all Chinese. Even US government websites have been found deploying Alibaba AI models. The adoption curve is steep and accelerating, driven by one factor above all else: price.

Alibaba’s Qwen, DeepSeek, and Moonshot AI’s Kimi are all open-weight models. That means their core parameters are publicly available for companies to download, modify, and run themselves. They underperform OpenAI and Anthropic on raw capability in some benchmarks, but at a fraction of the cost—which is precisely why enterprises are choosing them. DeepSeek’s R1 model, released in early 2025 at a dramatically lower price point than GPT-4-class alternatives, sent shockwaves through the industry and accelerated adoption across cost-sensitive segments. Financial services firms, healthcare startups, and mid-market SaaS providers—all categories that handle sensitive data but operate on thin margins—are among the fastest adopters.

Perhaps most significantly, many of these integrations occur beneath the surface. Chinese AI models are embedded in middleware, SDKs, and development frameworks that American engineers use without always tracking downstream dependencies. An app developer pulling a library from npm might inherit Qwen-based inference without realizing it. This opacity makes enforcement through disclosure inherently incomplete—the rule may catch the obvious cases while missing the deeper plumbing.

The Compliance Minefield

If this proposal becomes formal regulation, the operational consequences will be significant. Every mid-size software company, every cloud provider, every startup building on top of open-weight Chinese models will face a binary question: disclose or redesign your stack.

For global cloud providers operating in the US market, the compliance burden cuts both ways. A company running Chinese AI models on American servers faces scrutiny from Washington; a company refusing to host Chinese AI infrastructure risks losing business to competitors who do. The rule doesn’t distinguish between malicious dependence and pragmatic cost optimization—it demands disclosure of any and all Chinese AI usage, regardless of intent or risk profile.

Lawyers and compliance officers will spend months untangling the definitions. What counts as “using” a Chinese model? Is fine-tuning Qwen differently from running it directly? Does hosting an API endpoint that routes queries to Kimi constitute disclosure-worthy usage? The granularity of these questions will determine whether the rule bites broadly or narrowly. Internal audit teams will likely need to map entire supply chains—from model weights downloaded from GitHub repositories to inference endpoints spinning up on managed cloud instances—creating an unprecedented surveillance apparatus within private companies.

Smaller firms face disproportionate risk. Startups operating on lean budgets lack the legal infrastructure to navigate ambiguous compliance regimes. A misstep—whether an honest oversight or an over-cautious interpretation—could attract regulatory attention or negative publicity. The chilling effect may extend beyond disclosure: companies might preemptively abandon Chinese AI entirely, even in contexts where no security concern exists, simply to avoid the reputational hazard of being named.

Korea’s Uncomfortable Position

For South Korean semiconductor companies, the stakes cut deeper than mere compliance headaches. Korea’s memory chip and advanced packaging industries depend heavily on Chinese demand—particularly for AI-related chips sold to Chinese cloud providers and AI companies. Samsung Electronics and SK Hynix generate substantial revenue from the Chinese market. If the US disclosure rule forces American companies to visibly reduce their Chinese AI footprint, downstream demand for Chinese AI hardware could slow significantly, creating a revenue gap that Korean fabs would feel immediately.

More immediately, Korean companies that serve as middleware between US clients and Chinese AI models face direct exposure. A Samsung SDS subsidiary running Qwen on behalf of an American firm, for example, could find itself caught in the crossfire of competing regulatory regimes. Washington expects transparency; Beijing may interpret mandatory disclosure as hostile cooperation with US containment policy. South Korean firms have historically thrived by maintaining equilibrium between rival powers, but this proposal threatens to collapse that balancing act into a choice between two sovereign jurisdictions demanding loyalty.

Korea’s broader technology relationship adds another layer. The country sits at the intersection of three competing tech ecosystems: American closed-source giants, Chinese open-weight models, and Japanese and European players positioning themselves as neutral alternatives. Any US move toward mandatory disclosure pushes Korean firms to pick lanes in a geopolitical landscape where hedging used to be possible. Analysts at Seoul-based research firms have noted that Chinese AI companies are increasingly sourcing advanced packaging capacity from Korean suppliers—a symbiotic relationship that the disclosure rule could inadvertently disrupt by reducing the commercial viability of those Chinese AI developers.

Japan and the EU are watching closely. Japan’s METI has signaled interest in developing its own guidelines for AI supply chain transparency, while European Commission officials have privately acknowledged concerns about extraterritorial US rules affecting European companies operating in the American market. The Korean position—caught between compliance obligations and competitive survival—may become a template for other allied nations facing similar pressures.

What Happens Next

China’s diplomatic response was immediate and pointed. The Chinese embassy in Washington called the proposal harmful to global AI progress and vowed to protect the legitimate rights of Chinese companies. That language signals Beijing views this as more than bureaucratic friction—it sees an attack on its technology sector’s commercial viability. Chinese state media have framed the proposal as proof that America fears Chinese technological competitiveness, while diaspora tech communities in Silicon Valley have organized informal resistance efforts, sharing compliance guidance and warning members about potential exposure.

Inside the Trump administration, support for the proposal remains uncertain. The Reuters report explicitly notes that it is unclear how much backing the measure has within the executive branch or whether it will become actual regulation. The timing is notable, however. The discussion emerges alongside heightened scrutiny of Nvidia CEO Jensen Huang over his trips to China—a signal that the administration’s China-tech posture is hardening across multiple fronts simultaneously. Treasury Secretary Scott Bessent has faced repeated questioning from Congress about the scope of US financial exposure to Chinese AI firms, adding institutional momentum to the disclosure effort even as its ultimate fate remains undecided.

Industry groups including the Software & Information Industry Association and the Computing Technology Industry Association have begun mobilizing, drafting position papers that argue disclosure requirements are unworkable given the distributed nature of modern AI development. Their arguments may gain traction if the proposal proceeds to formal rulemaking, where notice-and-comment periods could surface additional technical objections.

The real test will come when the first company is forced to publish. Until then, the disclosure rule exists primarily as leverage—a threat hanging over boardrooms and general counsel offices, shaping behavior even without being formally enacted. That alone may achieve what a direct ban would not: chilling investment in Chinese AI before it reaches critical mass. Companies currently evaluating whether to adopt Chinese models for new projects may pause, reconsider, or choose American alternatives—not because they’re compelled to, but because the emerging regulatory climate makes the decision less clear-cut. In that sense, the proposal functions as policy before it functions as law, altering corporate behavior through expectation rather than enforcement. Whether that indirect pressure proves sufficient, or whether legal challenges and industry pushback prevent it from ever crystallizing into rule, will define the next chapter in the ongoing tech cold war.