US lawmakers examine corporate use of Chinese AI models

U.S. lawmakers are weighing ways to limit American companies’ use of artificial intelligence models developed in China, a review that could affect technology costs and procurement choices for firms serving the federal government. The scrutiny follows wider corporate interest in lower-cost Chinese models that CNBC reported have narrowed the performance gap with U.S. competitors.

The House Committee on Homeland Security and the House Select Committee on China opened a joint investigation in April into the use of Chinese-developed AI systems by U.S. companies. As an initial step, the committees sent letters to Cursor and Airbnb seeking information about their exposure to risks tied to China-built AI, according to CNBC.

AI has become a central point of competition between Washington and Beijing. In April, the Trump administration accused Chinese entities of conducting “industrial-scale campaigns” to copy U.S. AI systems and said it would examine ways to hold foreign actors accountable. Reuters reported Tuesday that Beijing is considering limits on overseas access to China’s leading AI models.

A State Department spokesperson told CNBC that the increased use of Chinese AI models by American companies “raises serious concerns,” saying those systems are designed to promote Beijing’s narratives, suppress dissent and reflect Chinese Communist Party ideology and values. A spokesperson for China’s embassy in the U.K. told CNBC that China “opposes baseless allegations and malicious smears against its AI development,” adding that the country’s AI sector rests on self-reliance and scientific and technological strength.

Cost and access drive adoption

Some U.S. government departments have barred Chinese AI models such as DeepSeek, but CNBC reported that American companies are not broadly prohibited from using them. Executives including Coinbase Chief Executive Brian Armstrong and Lindy founder Flo Crivello have publicly discussed using Chinese models to cut costs, according to CNBC.

Cursor built its Composer 2 model using Kimi, a Chinese AI model developed by Moonshot AI, CNBC reported. Cursor, which CNBC has reported is set to be acquired by Elon Musk’s SpaceX for $60 billion, declined to comment on the House probe.

Airbnb told CNBC that its AI work runs “overwhelmingly on U.S.-origin models.” The company said it uses a limited number of China-origin models, all open source, through approved U.S.-based service providers, with data and operations kept separate and protected.

Open-weight AI models add complexity for policymakers. In such systems, developers release model weights, the numerical parameters that help determine how the model produces outputs. Once those files circulate online, restricting access is harder than controlling a cloud service or a proprietary product.

Policy options under review

Andrew Garbarino, chairman of the House Committee on Homeland Security, told CNBC that the Chinese Communist Party is closing the gap in capabilities that will shape cybersecurity. He said reporting that a Chinese open-weight model can match leading U.S. models in some vulnerability discovery and cybersecurity tasks is “highly alarming.”

A committee aide told CNBC that lawmakers are also examining whether the United States has an adequate open-weight AI strategy, so American companies and cyber defenders are not left choosing between costly or restricted U.S. models and cheaper, capable alternatives from China.

Andy Ogles, chairman of the Subcommittee on Cybersecurity and Infrastructure Protection, called in June for a “serious strategy” to make American models a practical alternative. He said that if cheap and capable Chinese models become the easy choice, much of the world may build on them.

Kyle Chan, a fellow at the Brookings Institution’s John L. Thornton China Center, told CNBC that the administration could consider federal procurement bans, including limits on agencies and government contractors using Chinese AI models. Chan also said a full ban on open-source Chinese models would be difficult because model weights are freely available online and restrictions could raise First Amendment issues.

Daniel Remler, a senior fellow at the Center for a New American Security, told CNBC that U.S. officials appear concerned about corporate adoption but may hesitate if limits hurt startups or weaken support for open models. He said policymakers could use procurement rules or share risk and vulnerability findings with companies, and he expects both Congress and the executive branch to signal opposition to broad U.S. corporate reliance on the models.

This story draws on original reporting from CNBC.