Published November 21, 2026 · By
Chinese AI models are already woven into the software that many Americans use every day, and two House committee chairmen said Friday that the federal government should start pulling them out of its own systems. In a joint investigative report released Nov. 20, Rep. Andrew Garbarino (R-N.Y.), who chairs the House Homeland Security Committee, and Rep. John Moolenaar (R-Mich.), who chairs the House Select Committee on the Chinese Communist Party, laid out 12 recommendations aimed at limiting the reach of China-developed artificial intelligence in government work and critical infrastructure.
The report follows a seven-month inquiry into how U.S. companies, including Airbnb, Cursor maker Anysphere and DoorDash, have adopted open-weight Chinese AI models, which anyone can download, run and modify. Its recommendations are not law, and no bill was introduced alongside it.

What the House report on Chinese AI models found
The two committees opened their investigation in late April with letters to Airbnb and Anysphere, then added DoorDash in July. According to the committees, the work included document requests, company briefings, classified briefings and consultations with federal agencies. The Record, a cybersecurity news outlet, described the final product as a 50-page report.
The chairmen listed five main findings. First, they said Chinese AI models are already widely used across the U.S. economy, but much of that use is hard to track. Second, they concluded that low cost, easy customization and near-frontier performance drive adoption, and that the United States lacks a competitive open-weight alternative at the price and performance the market wants.
Third, the report argues that much of the apparent progress behind Chinese AI models rests on capabilities extracted from American models through a process known as distillation. Fourth, it says Beijing uses open-weight distribution, state support and low prices to deepen global dependence. Fifth, it warns that risks grow as the systems become more capable, because their behavior can differ in ways ordinary testing may not catch.
Which companies use Chinese AI models
The committees said Airbnb uses models from Alibaba in its customer support operations. Anysphere built the flagship model inside its Cursor coding tool on a base model from Moonshot AI, according to the report. DoorDash told investigators that several Chinese model families had been authorized and deployed inside the company.
The report also raises military concerns. It says personnel from China's People's Liberation Army cyber force visited the Beijing headquarters of AI developer Zhipu several times, most recently in August, and that a July procurement for a PLA cyber training range required integrating DeepSeek for what it called intelligent attack and penetration work.
Anadolu Agency, which reviewed the document, noted that the report did not establish that the use of Chinese AI models by any U.S. company had caused a cyberattack or a data leak.
The 12 recommendations on Chinese AI models
The chairmen's proposals mix restrictions with efforts to strengthen U.S. alternatives. According to the Select Committee's summary, they call on the government to:
- Launch a national American open-weight AI initiative and make trusted U.S. AI the global default.
- Create a model provenance standard and an "AI bill of materials" so buyers know where a model came from.
- Extend federal restrictions beyond DeepSeek and keep China-developed systems out of federal contract work and high-risk critical infrastructure functions.
- Bar federal cybersecurity grant money from being used to buy Chinese AI models, according to the Homeland Security Committee.
- Build long-term information-sharing channels with AI developers and test Chinese AI models systematically.
- Prepare the Department of Homeland Security and its cyber agency, CISA, to defend networks at machine speed and make CISA the single point for AI-related incident reports, with an independent board to investigate major incidents.
- Hold illicit distillers accountable, examine sales of American training data to Chinese labs, govern both open and closed models without overregulating, and coordinate with allies.
The chairmen describe the package as a way to reduce reliance on foreign systems while helping U.S. developers compete on price and performance. Several of the steps, such as an open-weight initiative or new grant restrictions, would require funding decisions or legislation from Congress.
The full release is posted on the House Homeland Security Committee website.
What each side has said
In a joint foreword, Garbarino and Moolenaar wrote that the risks are ones that businesses and officials "can no longer afford to overlook or downplay." They argued that Washington still has leverage over the issue and urged action now rather than years from now, comparing the spread of Chinese AI models to the long and costly effort to strip Huawei and ZTE equipment from U.S. networks. Summarizing that comparison, The Record said removing such models after deployment could take several years and cost taxpayers billions.
The committees said Airbnb and Anysphere have since engaged with them on the technical, commercial and security issues involved. The Record reported that none of the companies immediately responded to its requests for comment, and no response from the Chinese government was included in the coverage reviewed for this article.
No public statement from Democratic committee leaders on the new report appeared in the coverage reviewed. Concern about Chinese AI in federal systems has drawn support from both parties before, however. In June 2025, Moolenaar and Rep. Raja Krishnamoorthi (D-Ill.) introduced the No Adversarial AI Act, which would bar federal agencies from buying AI models developed by foreign adversaries, with Sens. Rick Scott (R-Fla.) and Gary Peters (D-Mich.) backing a Senate version.
Outside views
Mark Montgomery, a cybersecurity specialist at the Foundation for Defense of Democracies, told The Record the report highlights the risk of the United States not competing seriously in the market for low-cost open-source models. He added that companies may set aside security concerns in pursuit of a lower up-front cost, which matches the report's finding that low cost is a key reason Chinese AI models are spreading.
The report lands the same week President Trump said he and Chinese President Xi Jinping agreed to pursue joint AI guardrails, a pledge that has so far come with few details. It also adds to a broader debate in Washington over technology trade with Beijing, including scrutiny of U.S. AI chip sales to China.
What Chinese AI models mean for you
For most people, the report will not change how apps work tomorrow, and nothing in it bans consumers from using services built on Chinese AI models. Its recommendations target federal systems, government contractors and critical infrastructure, and they would need action by Congress or federal agencies to take effect. Here is what to watch, based on the report and our own analysis:
- Everyday apps: The report shows that Chinese AI models can sit behind familiar services, such as customer support chats, without users knowing. If you are concerned, check a company's privacy policy for how your data is processed and stored.
- Taxpayer costs: The chairmen argue that acting early is cheaper than a later removal effort like the Huawei and ZTE replacement program. Any new rules for agencies and contractors could carry compliance costs.
- Jobs and contractors: Businesses that sell software to the government may face new disclosure or sourcing rules if lawmakers turn the recommendations into legislation.
- Prices: The report itself says Chinese open-weight systems are popular partly because they are cheap. Our analysis: if companies replace Chinese AI models with U.S. alternatives, some could face higher AI costs, though the report gives no estimate.
The next step is up to Congress. The committees said they will continue their oversight, and any ban on Chinese AI models in federal systems would likely require new legislation or agency action in the new Congress that begins in January.
This article is for information only and is not financial or investment advice.
Sources: House Committee on Homeland Security (Nov. 20, 2026); House Select Committee on the CCP (Nov. 20, 2026); The Record by Recorded Future News (Nov. 20, 2026); Anadolu Agency (Nov. 21, 2026).