US officials are discussing a requirement that American companies publicly identify when they use Chinese artificial-intelligence models. The idea, reported by Reuters on November 18, would make model provenance visible to customers without immediately banning the technology. It is still a discussion inside the administration, not a regulation businesses have been ordered to follow.

What is being considered

Reuters reported, citing three people familiar with the discussions, that the US Treasury Department and White House Office of Science and Technology Policy have considered requiring disclosures on company websites. The report names Airbnb as an example of the kind of company that could be affected because of its use of Alibaba's Qwen technology in an AI customer-service agent. Airbnb declined to comment to Reuters; the agencies did not immediately answer the news agency's questions.

The critical qualification is that Reuters could not establish whether the proposal has enough support to advance, or whether a formal rulemaking process would follow. There is no published compliance deadline, final text, agreed definition of covered systems or confirmed enforcement mechanism. Describing the discussion as a new legal obligation would be premature.

Why Chinese models have become a business question

Developers can download and adapt open-weight systems such as Qwen, DeepSeek and Moonshot AI's Kimi, sometimes deploying them on infrastructure they control. Lower licensing and inference costs can make these models commercially attractive. A company selecting software for a support agent or coding assistant may care about accuracy, response speed, cost and integration more immediately than the nationality of the laboratory that originally trained it.

That is also why a blanket label could be misleading without additional detail. Running open weights on an American cloud does not necessarily send every customer conversation to China. Conversely, hosting a model domestically does not remove concerns about its software supply chain, training provenance, update process or behavior. Any disclosure regime would have to distinguish model origin, hosting location, data access and the application in which the system operates.

Washington's concern is strategic dependency. The Reuters report says US officials are considering publicity as a way to discourage reliance on Chinese suppliers while avoiding a direct prohibition. It also describes congressional scrutiny of companies using Chinese models. Those concerns do not, by themselves, establish that a particular deployment leaked private information or violated US law.

Separate the idea from existing chip controls

Earlier this week, the US government announced a separate change to controls on advanced AI chips and remote computing access. That measure addresses the availability of certain hardware capabilities and customers across borders. The new model-disclosure discussion concerns software choices made by American businesses. Confusing them would make the policy appear much broader—and more settled—than it is.

Competition also complicates the story. Open-weight models have allowed smaller firms to build and customize products without paying for access to a single proprietary API. Nextwith.ai recently covered Alibaba's newer Qwen model and its unverified safety and cost claims. Whether a customer chooses that ecosystem should still depend on documented task performance, data handling, security review and legal requirements—not a country label alone.

What businesses should watch

The next test is whether the administration publishes a formal proposal with precise definitions. It would need to answer whether a disclosure applies to model weights used entirely on private infrastructure, to services that route only some tasks through a Chinese model, or to applications built on models subsequently fine-tuned by US developers. It would also need a process for updating the disclosure when software is replaced.

Companies can already maintain an internal AI inventory recording the model, supplier, host, permitted data and use case. That is a sensible governance practice independent of any hypothetical new rule. What they should not do is present the Reuters account as proof that public disclosure is now mandatory. As of November 18, the development is an early policy discussion whose final form—and existence—remain uncertain.

Sources