The Treasury Department and the White House Office of Science and Technology Policy have discussed requiring US companies to disclose their use of Chinese AI models on their websites, three people told Reuters in a report published on 18 November. The administration is considering disclosure as a way to discourage reliance on the models without imposing restrictions on their use.
Key points
- Companies such as Airbnb would have to post their use of Chinese AI models on their websites under the plan being discussed.
- House committee chairs have sought information from Airbnb, Anysphere and DoorDash about their use of Chinese models.
- The proposal’s standing within the administration and its possible route through rulemaking were not immediately clear.
Treasury and White House officials discuss website notices
Under the plan described by the three people, a company using a Chinese model would have to make that use public on its own website. The proposed obligation concerns disclosure rather than approval of a model or a ban on its deployment. Reuters reported that it was not immediately clear whether the idea was gaining support within the administration or would enter a formal rulemaking process if taken forward.
The discussion centres on open-weight models, whose core components can be downloaded and adapted. Chinese-made models including Moonshot AI’s Kimi, DeepSeek and Alibaba’s Qwen are gaining ground with US companies because they cost less than proprietary tools from OpenAI and Anthropic and perform well enough, according to the Reuters report. On OpenRouter, an AI marketplace that tracks usage, Chinese models occupied the top three places by tokens processed in October. Tokens are the units of data handled by a model.
A website notice would leave companies able to choose those models but put the choice in public view. John Moolenaar, who chairs the House select committee on China, welcomed the deliberations. He said that disclosure was an important step in managing potential risks from Chinese open-weight models, including when US companies run them on their own infrastructure.
Airbnb, Anysphere and DoorDash face congressional questions
Airbnb uses Alibaba’s Qwen model for its AI customer service agent. In a statement, the company said its AI activity “runs overwhelmingly on U.S.-origin models”. It said it uses “only a limited number of China-origin models”, which it described as open-source and operated through approved US-based service providers, with data and operations kept separate.
Republican Representatives Andrew Garbarino of New York and Moolenaar of Michigan began investigating American companies’ use of Chinese AI this year. Garbarino chairs the House Homeland Security Committee. The two lawmakers sent letters to Airbnb, Anysphere and DoorDash. Anysphere makes the AI coding tool Cursor, and its Composer 2 model was built on a version of Kimi. A DoorDash co-founder has said the company routes some tasks in an internal system through Kimi.
The inquiry covers distinct kinds of use: a customer-facing agent at Airbnb, a model built on Kimi at Anysphere, and tasks within DoorDash’s internal system. The Chinese Embassy in Washington said Beijing opposes politicising trade and technology issues and would defend Chinese businesses’ lawful interests.
Bessent’s sanctions warning and the chip rule
Treasury Secretary Scott Bessent said in July that the administration could sanction Chinese AI companies if they were found guilty of intellectual property theft through distillation of American models. Distillation trains a smaller model using outputs from a larger one. Michael Kratsios, director of the White House Office of Science and Technology Policy, accused China in April of taking US AI laboratories’ intellectual property on an “industrial scale”. US authorities have so far refrained from curbs or penalties aimed at Chinese AI companies over their models.
The administration has taken a different approach to the hardware used to run AI. On 16 November, the Commerce Department’s Bureau of Industry and Security issued an interim final rule requiring special licences for certain advanced AI chip sales to data-centre customers in Indonesia, Malaysia, Singapore, Thailand and Brazil, NBC News reported. Existing overseas data-centre customers must also obtain a licence if they provide significant computing capacity to companies headquartered in countries barred from receiving certain advanced US technology, including China. AI Affairs previously reported that Chinese AI firms sought overseas Nvidia capacity ahead of possible US curbs.
Chinese models are also in use within the US government. A federal website deployed an Alibaba AI model so users could search proposed regulations, Reuters reported in September. The deployment came despite an FBI accusation that the model’s developer had copied Anthropic’s technology.