US lawmakers warn American firms risk Chinese AI dependence

Share:
Audio Loading voice…
US lawmakers warn American firms risk Chinese AI dependence

Synopsis

A US congressional report has revealed that Airbnb handles 80,000 daily exchanges using Chinese AI, DoorDash would need six months to replace one embedded model, and Anysphere quietly built its flagship product on a Chinese base model — while the same AI families appear in PLA military procurement. The finding reframes Chinese open-weight AI not as a cost-saving tool but as a potential structural dependency inside American critical infrastructure.

Key Takeaways

Airbnb uses Alibaba's Qwen models to handle ~80,000 exchanges daily and summarise ~250,000 support cases per day.
DoorDash has deployed 25 Chinese model families and estimates it could take six months to replace one deeply embedded model.
Anysphere's Cursor flagship product, Composer 2 , was built on a base model from China's Moonshot AI — disclosed only after independent discovery.
A PLA cyber training facility tender from July required integration of DeepSeek ; some of the same model families are in use at major US firms.
House committee chairmen John Moolenaar and Andrew Garbarino proposed an 'AI bill of materials' and called for Chinese models to be banned from federal contracts and critical infrastructure.
Major Chinese developers allegedly used fraudulent accounts and proxy networks to harvest outputs from leading American AI systems, according to federal investigations cited in the report.

Two powerful US House committees released a report on Friday, 21 November 2026, warning that American companies are becoming dangerously reliant on Chinese artificial intelligence models — citing evidence from Airbnb, DoorDash, and coding platform maker Anysphere. The report flagged that the same model families used in major US corporations also appear in People's Liberation Army (PLA) procurement documents.

Key Findings From the Report

Airbnb told investigators it runs customised versions of Alibaba's Qwen models on US-based servers. One model powers an AI assistant handling approximately 80,000 exchanges a day with guests and hosts; another produces summaries of roughly 250,000 customer support cases and calls daily. DoorDash, meanwhile, disclosed that its engineers had downloaded or evaluated models from at least six Chinese developers, deploying 25 Chinese model families whose computing and operating costs reached at least $5,000 each. The delivery giant estimated it could take as long as six months to replace one deeply embedded model.

Perhaps the most striking disclosure involved Anysphere, maker of the Cursor coding platform. The company built its flagship Composer 2 model on a base model from China's Moonshot AI — a detail that only became public after an independent developer identified the underlying model, according to the report.

The Chinese Military Connection

The committees examined Chinese military procurement records and found that a July tender for a PLA cyber training facility required the integration of DeepSeek. A separate military project called for models from four Chinese developers to operate on an internet-isolated network. The committees noted that some of the same model families flagged in military procurement records were simultaneously in active use at major American corporations — a convergence that the chairmen described as a significant national security concern.

What the Committees Are Calling For

John Moolenaar, who chairs the House Select Committee on China, and Andrew Garbarino, who chairs the House Homeland Security Committee, called for a national effort to develop American open-weight models that companies can download and adapt. The two Republicans acknowledged that US developers had yet to offer an alternative matching the combination of price, performance, and flexibility that is drawing businesses to Chinese models.

The chairmen urged that Chinese-developed models be kept out of federal contract work and critical infrastructure operations. They also proposed an 'AI bill of materials' — a disclosure framework that would help buyers trace the origins of the models powering a given product.

Allegations of Data Theft and Proxy Networks

Citing federal investigations, the chairmen alleged that major Chinese developers had used fraudulent accounts and proxy networks to collect responses from leading American AI systems and reproduce some of their capabilities — essentially harvesting the outputs of US-built models to accelerate their own development. The committees also urged scrutiny of American firms that supply labelled training data to Chinese AI laboratories.

Notably, both Airbnb and DoorDash confirmed they had no commercial agreement with any Chinese model developer and had paid none of the developers. Running the models on US-based servers substantially reduces the risk of user data being sent directly to China, the committees acknowledged — but also means that spending figures reveal little about how widely the technology is actually used.

What Happens Next

'We should not wait five years to discover that much of the world, and much of the United States, runs on AI shaped by the CCP, leaving policymakers with little choice but another costly rip and replace effort,' the chairmen wrote, referring to the Chinese Communist Party. The committees opened their inquiry in April with letters to Airbnb and Anysphere, expanded it to DoorDash in July, and received documents and briefings from all three companies. Legislative proposals on federal procurement rules and the AI bill of materials framework are expected to follow.

Point of View

But it addresses disclosure, not dependency — and disclosure without a credible American alternative simply confirms the problem without solving it. Until US developers close the price-performance gap on open-weight models, the structural incentive for American firms will remain unchanged.
NationPress
21 Nov 2026

Frequently Asked Questions

Why are US lawmakers concerned about Chinese AI models in American companies?
A congressional report released on 21 November 2026 found that major US firms including Airbnb and DoorDash are deeply embedded with Chinese AI models, while some of the same model families appear in People's Liberation Army procurement records. The committees warned this creates a dependency that could leave American critical infrastructure shaped by technology linked to the Chinese Communist Party.
Which Chinese AI models are US companies using?
Airbnb runs customised versions of Alibaba's Qwen models. DoorDash has deployed 25 Chinese model families from at least six developers including DeepSeek. Anysphere built its Cursor coding platform's flagship Composer 2 model on a base model from China's Moonshot AI.
Does using Chinese AI models mean US user data is sent to China?
Not necessarily. Both Airbnb and DoorDash said they run the models on US-based servers, which substantially reduces the risk of user data being transferred to China, according to the committees. Neither company had a commercial agreement with any Chinese model developer.
What policy actions are the House committees proposing?
The committees proposed an 'AI bill of materials' to help buyers identify and trace the origins of models inside a product. They also called for Chinese-developed models to be excluded from federal contract work and high-risk operations in critical infrastructure, and urged a national effort to develop competitive American open-weight alternatives.
How did Chinese AI models become so widely used in US businesses?
According to the report, Chinese open-weight models offer a combination of price, performance, and flexibility that US developers have not yet matched. Because open-weight models can be downloaded and run on a company's own servers, businesses can adopt them without a commercial relationship with the developer, making adoption easy and spending figures an unreliable measure of how widely they are used.
Nation Press
The Trail

Connected Dots

Tracing the thread behind this story — newest first.

8 Dots
  1. Latest 20 min ago
  2. Yesterday
  3. 2 months ago
  4. 3 months ago
  5. 3 months ago
  6. 3 months ago
  7. 4 months ago
  8. 6 months ago
Google Prefer NP
On Google