U.S. lawmakers request information from DoorDash on use of Chinese AI

U.S. lawmakers have requested info from meals supply firm DoorDash on its use of Chinese language artificial intelligence fashions, CNBC has realized, as scrutiny round American companies’ use of methods developed by China ramps up.

In a letter obtained by CNBC, the chairmen of two Home committees conducting a joint investigation into safety implications of U.S. firms utilizing Chinese language AI fashions requested DoorDash to share “info and paperwork” referring to its analysis and deployment of AI methods from China.

“DoorDash proudly helps American AI management and is working to make sure AI advantages Essential Road, not simply the largest firms,” a DoorDash spokesperson informed CNBC. “We look ahead to partaking with the Committees on how we safely and responsibly use AI, together with American-developed frontier fashions and open-weight fashions.”

Rising adoption of China-built AI models has led to rising calls from U.S. lawmakers for methods to fight the pattern, together with by way of an ongoing investigation from The Home Committee on Homeland Safety and the Home Choose Committee on the Chinese language Communist Social gathering.

An preliminary step within the joint investigation was for the chairmen of these committees to ship letters to Cursor and Airbnb, over their “use of or publicity to those dangers” via AI developed in China.

The letter cited a post on X by DoorDash founder Andy Fang that particulars how the corporate is delegating lower-level AI work to Chinese language AI mannequin Kimi K2.6, which is developed by Moonshot AI.

DoorDash’s AI analysis lab had stated on X that it had seen Kimi K2.6 and Anthropic’s Fable 5 vastly outperform different Anthropic fashions it had used, together with “Sonnet 4.6 and Opus 4.8 harness at a less expensive value.”

“The Committees acknowledge that U.S. firms, from giant expertise companies to startups, could consider and deploy PRC-developed open-weight fashions as a result of they’ll present aggressive capabilities, decrease prices, better customization, and alternate options to reliance on a small variety of proprietary mannequin suppliers,” the letter reads.

It added: “These sensible concerns don’t eradicate the necessity for risk-based safeguards or diminish the nationwide safety considerations related to rising dependence on fashions developed by entities topic to PRC jurisdiction.”

AI has emerged as a key level of rivalry between the U.S. and China, with each nations vying for supremacy within the discipline.

“The Chinese language Communist Social gathering is not simply nipping at our heels in synthetic intelligence; it’s racing to shut the hole in among the actual capabilities that may form the way forward for cybersecurity,” Andrew Garbarino, chairman of the Home Committee on Homeland Safety, beforehand informed CNBC.

“Current reporting {that a} Chinese language open-weight mannequin can match main U.S. fashions in sure vulnerability discovery and cybersecurity duties is extremely alarming,” stated Garbarino.

Moonshot AI’s launch of open weight model Kimi K3 earlier this month claimed to have largely closed the efficiency hole with main U.S. fashions.

Whereas some authorities departments have banned the utilization of Chinese language AI fashions like DeepSeek, adoption by U.S. firms is just not prohibited. Tech chiefs, together with crypto firm Coinbase‘s Brian Armstrong and AI startup Lindy’s Flo Crivello, have been publicly touting using fashions from China to cut back prices.

“An efficient federal strategy ought to due to this fact scrutinize U.S. firms’ reliance on [People’s Republic of China]-developed fashions and strengthen the provision, safety, and competitiveness of American open-weight alternate options,” the letter stated.

The supply of open weight fashions was thrown into the highlight lately after it emerged {that a} cyber assault by rogue OpenAI fashions on Hugging Face was stopped by utilizing a Chinese system.

Open weight fashions could be downloaded, modified and self-hosted by firms. Essentially the most succesful open weight fashions are Chinese language made. The main frontier fashions developed by OpenAI and Anthropic are closed.

“The Committees are additionally analyzing whether or not the USA has a ample open-weight AI technique to make sure American firms and cyber defenders should not compelled to decide on between costly or restricted U.S. fashions and low cost, succesful PRC-developed alternate options,” a committee aide, who requested to not be named as they weren’t approved to debate the continued probe, beforehand informed CNBC.