US Lawmakers Question DoorDash Over Use of Chinese Artificial Intelligence Models

Chairmen from two key House committees have formally requested documentation from DoorDash concerning its utilization of AI systems engineered in China. The inquiry, conducted by the House Committee on Homeland Security and the House Select Committee on the Chinese Communist Party, seeks to understand how the delivery giant evaluates and deploys foreign software. This move follows previous outreach to other organizations, including Airbnb and Cursor, signaling a coordinated effort to scrutinize the growing trend of U.S. companies integrating Chinese technology into their operations. The investigation is rooted in a desire to understand the potential security implications of this reliance, as lawmakers aim to discern whether these business practices could inadvertently undermine domestic technological autonomy or expose infrastructure to vulnerabilities that could be exploited by foreign entities.

The congressional scrutiny follows a public acknowledgment by DoorDash founder Andy Fang, who shared via social media that the company utilizes the Kimi K2.6 model, developed by Beijing-based Moonshot AI, for specific low-level tasks. Internal research from the company’s AI laboratory reportedly suggested that Kimi K2.6 outperformed several models from Anthropic—specifically Sonnet 4.6 and Opus 4.8—while simultaneously offering a more cost-effective solution. A company spokesperson emphasized that DoorDash remains committed to American AI leadership and intends to cooperate with the committee. They stressed that the company’s internal policy is centered on the safe and responsible use of a diverse range of systems, which includes both domestically produced frontier models and various open-weight alternatives.

The U.S. government views the advancement of AI as a primary theater of competition with China, with lawmakers increasingly wary of the rapid progress of foreign-made systems. Rep. Andrew Garbarino, chairman of the House Committee on Homeland Security, has highlighted concerns regarding recent data suggesting that Chinese open-weight models have reached a level of sophistication comparable to, or in some tasks exceeding, leading American alternatives. This includes proficiency in cybersecurity-related functions and vulnerability discovery. While the federal government has placed restrictions on using specific systems, such as DeepSeek, in government departments, private companies remain largely uninhibited in their adoption. The committees are now debating whether current U.S. strategy sufficiently supports domestic open-weight development, or if American firms are being forced to turn to Chinese alternatives due to the high costs or restrictive licensing of U.S.-based frontier models.

The current investigation highlights a pivotal shift in the AI industry toward open-weight models, which companies can download, modify, and host on their own infrastructure. These models provide significant advantages in terms of customization and cost, making them attractive to businesses of all sizes, from tech giants to emerging startups. However, this portability also presents unique security challenges. The congressional interest is partly fueled by the realization that these Chinese systems are increasingly capable, with models like Moonshot AI's Kimi K3 claiming to have bridged the performance gap with top-tier American products. Because these models are not tethered to a single proprietary provider in the way that closed-source systems are, their proliferation across the private sector has complicated the task of creating a coherent national security policy regarding foreign-developed AI.