House Lawmakers Urge US Open-Weight AI Initiative to Counter China
A new report from the House Select Committee on China recommends a federal initiative to foster American-made open-weight AI models, aiming to reduce reliance on Chinese alternatives.

House lawmakers have called for the establishment of a U.S. initiative focused on developing American-made open-weight Artificial Intelligence models, a move intended to counter the growing influence of Chinese AI technologies within the United States. The recommendation comes from the House Select Committee on China, which published a report detailing the widespread adoption of Chinese open-weight models such as DeepSeek, Kimi, Qwen, and GLM by American companies.
The committee's investigation revealed that U.S. firms, including Airbnb, Anysphere, Cursor, and DoorDash, have increasingly turned to these Chinese models due to their perceived advantages in cost, ease of customization, and performance. However, the report highlights that much of this usage remains difficult to track, raising concerns about potential security risks and a lack of transparency.
Lawmakers expressed increasing concern that Chinese models may be producing less secure code and could be used to autonomously build cyber exploits. To address these risks, the report proposes a multi-pronged approach. Key recommendations include launching a national initiative that combines public computing resources, shared evaluation infrastructure, and lawful access to leading American AI models for training purposes, supported by federal procurement demand.
Furthermore, the committee urged the White House to prohibit federal agencies from utilizing Chinese AI models and to subject all such models to consistent, rigorous evaluation. The report also suggests sanctioning companies that engage in the illicit distillation of U.S.-made AI models, a practice where developers use proprietary models to train their own without proper authorization.
This push for domestic alternatives comes as some American companies are already moving away from Chinese models. For instance, Cursor, recently acquired by SpaceX, reportedly plans to discontinue its use of Chinese models, having previously built its coding model Composer 2 using Kimi K2.5 from Moonshot AI.
The report acknowledges the existence of American open-weight models, such as Meta's Llama and Muse, and Nvidia's Nemotron, as well as open-weight offerings from OpenAI and Google. However, it also notes the ongoing debate between proponents of proprietary and open-weight models, with some experts arguing that proprietary models remain more capable and secure. Despite this, the committee observed that Chinese open-weight models are rapidly improving in capability, potentially outpacing current safety evaluation methods.
Distillation, a technique where AI models are trained on the outputs of other advanced models, is cited as a potential reason for the rapid development of Chinese open-weight models. While accepted in some research contexts, the report, citing information from Anthropic and OpenAI, suggests that many Chinese labs employ this method illicitly, without permission, further exacerbating security concerns.