Skip to main content
Industry News
· 6 min read

Your Website Might Have to Disclose Its AI: The Chinese-Model Rule Washington Is Weighing

Flat-vector line-art illustration on a dark navy background: a small-business website browser window with a chat bubble and AI agent icon, a provenance checklist card and a US flag shield badge beside it, accented in electric blue

On November 18, 2026, Reuters reported that the Trump administration is considering requiring US companies to publicly disclose on their websites when they use Chinese AI models — a "name and shame" approach to discourage reliance on tools like Alibaba's Qwen, Moonshot AI's Kimi, and DeepSeek without imposing new restrictions. Two days later, on November 20, the chairmen of two House committees dropped a joint investigative report on the same models, carrying twelve policy recommendations — including a model provenance standard and an "AI bill of materials," plus calls to keep PRC-developed models out of federal systems and high-risk critical infrastructure. Both stories are about big companies. The version that matters for a Temecula or Murrieta small business is simpler: when your AI tools get cheaper than the American alternatives, you are about to be asked which side of that list your website is on. Here is what is actually on the table, how it might reach your site, and the five-line disclosure you can write today.

What Is Actually on the Table

The Reuters exclusive, by Alexandra Alper, is specific about mechanics: the Treasury Department and the White House Office of Science and Technology Policy have discussed a plan under which companies would be required to post their use of Chinese AI models on their websites. Airbnb is the named example — its AI customer service agent uses Qwen, an AI model made by China's Alibaba. It is not yet clear whether the idea is gaining traction inside the administration or would go through formal rulemaking; the White House, Treasury, and Alibaba all declined to comment, and an Airbnb statement said its AI activity "runs overwhelmingly on U.S.-origin models," with only a limited number of open-source China-origin models run through approved US-based service providers.

Why the sudden attention: Chinese-made open-weight models are cheaper than proprietary tools from OpenAI and Anthropic and perform well enough that Chinese AI models were the three most popular by token usage in October on OpenRouter, an AI marketplace that tracks usage data. Congress has been watching since April — Garbarino and Moolenaar's investigation sent letters to Airbnb, Anysphere (whose Cursor coding tool's Composer 2 model was built on a version of Kimi), and DoorDash (whose co-founder said some internal tasks route through Kimi). The November 20 report says plainly that PRC open-weight models are "already widely used across the American economy, but much of that use is hard to track," and that cost, customization, and near-frontier performance are driving adoption.

Two things it is not. It is not a ban — officials have stopped short of restricting or punishing Chinese AI companies, and the whole disclosure framing exists precisely because the administration wants to avoid new mandates. And it is not only about giants: the report's own answer to "hard to track" is a provenance standard and an AI bill of materials that would flow down through vendors and agencies to whoever sells to them.

Why This Matters Locally

  • The disclosure may land on the website you own. The mechanism under discussion is a public website posting, not a regulatory filing. If a rule or pressure campaign takes shape, the businesses that get asked "which AI runs your chatbot and phone agent" first will be the ones already using AI publicly — the chat widget, the missed-call texter, the review responder. Your site becomes the disclosure surface.
  • The cheap-model temptation is the whole story. Chinese open-weight models are winning on price — that is the explicit finding. A free or near-free model inside a $29 chatbot tool is exactly the kind of invisible dependency the report says is "hard to track." If your AI features came from a reseller, you may not know what model is underneath; the vendor's cost choice becomes your disclosure.
  • "I don't know" is the expensive answer. The report recommends a model provenance standard and an AI bill of materials precisely because buyers can't currently see what they're running. When a prospect, insurer, or agency contract asks, the business with a one-paragraph answer keeps the deal; the one without stalls it.
  • Federal and critical-infrastructure adjacency is real work. The report calls for extending the existing DeepSeek exclusion from federal civilian systems — banning PRC models from federal contract work and high-risk critical infrastructure functions. A Temecula contractor or vendor selling into any federal-adjacent pipeline should expect model-origin questions on intake forms.
  • Vendor risk has been the pattern all year. This is the same shape as Character.AI's bankruptcy and the arrival of AI liability insurance: your exposure is decided by vendors' choices, not yours. The DeepSeek agent wave we covered when it launched is now the exact category Washington wants named on websites.

The Five-Step Playbook

  • Inventory what AI you actually run. List every AI touchpoint — chatbot, phone agent, review responder, email drafter, site builder's AI features — and for each, get from the vendor, in writing, the model family and origin behind it. One afternoon of asking beats a season of guessing.
  • Write your one-paragraph AI disclosure now, voluntarily. Name the tools, what they do, where customer data goes, and that a human reviews consequential work. If disclosure ever becomes expected or required, you already meet it; until then it's a trust signal competitors won't have.
  • Set a simple origin policy before you shop. Something like: US-hosted models for anything that touches customer data or speaks for the business; open-weight experiments only on internal, low-stakes work. Choosing your answer before the pressure arrives is cheaper than improvising one.
  • Put the disclosure on your website, in plain English. The mechanism under discussion is a website posting — a short "How we use AI" page or a footer-linked section is exactly the shape it would take. Vague marketing copy ("powered by AI") is what this movement is designed to replace.
  • Ask the provenance question in every AI contract. Model family, hosting location, data handling, and what happens if the vendor swaps models. That's the AI bill of materials, shrunk to one email — and it's the question your vendors' vendors are already being asked.

The Bottom Line

Washington has found its AI policy tool and it is not a mandate — it's a spotlight pointed at websites. Whether or not the disclosure idea survives contact with rulemaking, the direction is set: cheap Chinese open-weight models are winning on price, Congress wants them named, and the asking will flow downhill to small businesses through vendors, insurers, and agency contracts. The cheap move now is the one-paragraph inventory-plus-disclosure; the expensive one is discovering your chatbot's model origin from a prospect's intake form. When you want AI employees with an honest label on the box — US-vendor AI employees from $397/mo, websites from $1,497 one-time — that's what we build. Free setup, no contracts, cancel anytime.

Sources