A recent investigation conducted by two committees in the United States House of Representatives revealed significant concerns over the extensive infiltration of Chinese “open-weight artificial intelligence (AI) models” into American businesses. The investigation, which culminated in a joint report, highlighted the potential risks these models pose in terms of deepening technological dependency and threatening national security.
The two committee chairs have raised alarms over the need to curb the use of Chinese-developed models in federal systems and critical infrastructure. There is a pressing call for support of domestic AI models to mitigate these risks.
“Open-weight AI models” refer to systems where the core components are publicly accessible, allowing users to download, modify, and tailor the model for specific tasks. This open accessibility has attracted American companies seeking affordable and customizable solutions that deliver near state-of-the-art performance.
The 50-page report encapsulates findings from approximately seven months of investigation, examining American firms such as Airbnb, DoorDash, and Cursor. It notes that the appeal of Chinese models lies in their low cost and adaptability. However, once these models are downloaded, modified, and integrated into business systems, tracing their origins becomes challenging. This complicates future replacements and could potentially escalate operational risks and costs.
Andrew Garbarino, chair of the House Homeland Security Committee, and John Moolenaar, chair of the Select Committee on the Chinese Communist Party, emphasized that the investigation unveiled risks that American businesses, policymakers, and national security officials can no longer afford to ignore or downplay.
Their recommendations are clear: there should be a prohibition on the use of AI models developed in China, along with their derivatives, in federal information systems, federal contracts, and critical infrastructure that could potentially cause physical harm or disrupt essential services. They did suggest a limited exemption for research, security testing, and scenarios where there are no credible alternatives available.
Furthermore, the report encourages the U.S. government to provide computational resources, testing facilities, and procurement support to bolster domestic open-weight AI model development. It also advocates for leading American AI companies to voluntarily license developments to domestic developers for training purposes.
In prior actions, the U.S. Department of Homeland Security has already issued an order prohibiting federal civilian agencies from procuring products and services from DeepSeek, while demanding the removal of its technologies from government information systems.
Notably, the report indicates that Airbnb utilizes Alibaba’s Qwen model in its service operations within the United States. This AI assistant currently processes approximately 80,000 interactions daily between guests and hosts. In addition, another model is tasked with summarizing around 250,000 customer service inquiries and calls each day.
DoorDash disclosed to Congress that it has approved and deployed a series of 25 Chinese AI models. Among them, the Kimi K2.6 model from “The Dark Side of the Moon” is utilized for preliminary code reviews. The company acknowledged that most of these models could be replaced swiftly; however, one model that is entrenched within a larger system might take up to six months to substitute.
Cursor, a subsidiary of SpaceX, bases its Composer 2 model on the Chinese Kimi K2.5. Nonetheless, the company informed Congress that due to “geopolitical risks,” it is considering gradually phasing out the use of Chinese open-weight models.
Both Airbnb and DoorDash confirmed that their relevant models operate within U.S. borders. The Congressional report asserts that this significantly reduces the risk of user data being transmitted directly to China.
In its appendix, the investigation report noted that individuals associated with the Chinese Communist Party’s cyber operations have frequently appeared at the headquarters of the Chinese AI firm Zhipu, with the most recent visit occurring in August of this year.
The committee assessed with “moderate confidence” that these contacts indicate a potential channel between Zhipu and the military intelligence operations of the Chinese Communist Party.
The committee uncovered procurement records from the Chinese military, revealing that in 2026, a cyber target acquisition involved the integration of DeepSeek to facilitate “smart attacks” and “intelligent infiltration.” Another procurement required the operation of DeepSeek, Qwen, Kimi, and MiniMax models on an isolated intranet to develop military AI agents and software. These records showcase military procurement demands, although not all projects may have reached deployment.
The report also provides a comparative analysis of open API pricing, indicating that the Alibaba Qwen model charges approximately $1.15 for every million input and output tokens, while OpenAI’s GPT-5.5 charges around $35. However, it is essential to note that this comparison did not account for differences in model capabilities and deployment costs.
Additionally, data cited from Hugging Face suggests that Chinese models accounted for 41% of downloads on their platform over the previous year.
The report references warnings from entities such as the U.S. National Security Agency (NSA) and the Federal Bureau of Investigation (FBI), indicating that some Chinese developers exploit fake accounts and proxy networks to harvest output content from American leading models, engaging in unauthorized model distillation.
It highlights ongoing support from Chinese authorities for domestic model development through national investments and computational subsidies.
Mark Montgomery, a cybersecurity expert at the Foundation for Defense of Democracies, pointed out that insufficient competition within the low-cost open-source model market in the U.S. could lead to security vulnerabilities.
Montgomery remarked, “Enterprises and utility companies, based on considerations of cost and resources, will continue to prioritize reducing upfront costs while neglecting safety and protection issues. The Chinese side is acutely aware of this modus operandi.”