MAIN CONTENT
Disclosure as a policy tool, not a ban
Three people familiar with the discussions told Reuters that American businesses could be asked to disclose their use of Chinese AI models on their own websites. Such an approach would place the choice before customers, investors and business partners rather than prohibit adoption directly. It would differ from export controls and other restrictions used by Washington to limit transfers of advanced technology to China.
Yet the government has not published an enforceable requirement, scope or timetable. A future proposal would have to define whether disclosure covers experiments, internal tools, cloud-hosted applications and customized models. It would also need to protect legitimate commercial secrets while providing useful information to customers. For now, describing the discussions as an existing legal duty would overstate what is publicly known.
Why American companies consider Chinese models
Alibaba’s Qwen, Moonshot AI’s Kimi and models associated with DeepSeek have gained attention because their capabilities can satisfy practical needs at competitive cost. Open-weight releases let developers download model parameters and tailor or host systems within their own environments.
This is not the same arrangement as transferring enterprise data to a server controlled by the original developer in China.
Airbnb told Reuters that its AI activity runs overwhelmingly on US-origin models. It described its use of Chinese-origin systems as limited, open-source and managed exclusively through approved US-based service providers to keep data and operations protected. The distinction matters: in purchasing AI, a company must consider the model, the host, data retention, security testing, price and contract terms, rather than make a decision on nationality alone.
Real security questions and political perceptions
Open-weight models can expose companies to vulnerabilities in supporting software, model supply chains, modifications and downstream applications. But the country where a model was developed is not, by itself, proof that sensitive information is being sent to its creator. A meaningful assessment should examine model versions, permissions, data paths, audit logs and organizational controls.
Representatives Andrew Garbarino and John Moolenaar have investigated corporate use of Chinese AI, including letters involving Airbnb and companies behind tools such as Cursor. Moolenaar welcomed disclosure as a way to manage perceived risks even when models run on domestic infrastructure. The Chinese Embassy, for its part, objected to politicizing trade and technology and argued that doing so would hamper global AI progress. These positions highlight a dispute over evidence, competition and trust.
The choices facing regulators and industry
US officials face competing objectives: protecting critical capabilities while avoiding unnecessary barriers to innovation. Clear disclosures might help stakeholders understand which AI systems influence automated decisions. Poorly designed notices, however, could imply that every model of one origin carries identical risks, whatever the operational safeguards.
A stronger governance approach would test security controls, contractual accountability, hosting arrangements and incident response across providers. Research from CSIS has emphasized that Chinese open-weight adoption is also a competition over developer ecosystems, deployment cost and lasting technological influence. Effective policy therefore needs measurable standards as well as clear definitions, rather than reliance on a country-of-origin label alone.
















