DoorDash faces congressional scrutiny over Kimi AI deployment

House Select Committee on China Chairman John Moolenaar and Homeland Security Committee Chairman Andrew Garbarino sent DoorDash co-founder and chief executive Tony Xu a letter on Friday requesting information and documents on how the company evaluates and deploys Chinese-developed AI.

The request grew out of a post on X by DoorDash founder Andy Fang describing the delegation of lower-level AI work to Kimi K2.6, built by Beijing-based Moonshot AI, per CNBC.

DoorDash is not accused of breaking any law. US companies may freely run Chinese models, though federal agencies have blocked some systems including DeepSeek on government devices.

A company spokesperson said DoorDash “proudly supports American AI leadership” and looks forward to engaging with the committees.

Lawmakers say US model pricing is part of the problem

The chairmen conceded the business case in their own text. American firms may deploy Chinese open-weight models because they deliver “competitive capabilities, lower costs, greater customization,” the letter says, and because they offer an alternative to depending on a small number of proprietary providers.

Those considerations, the chairmen wrote, do not remove the need for risk-based safeguards.

The committees claim that “US policies need to ensure that open-weight designs offered by the US become more accessible, safer, and competitive for businesses to discourage them from using Chinese options.”

The committee’s larger argument here is quite straightforward: American businesses and cybersecurity practitioners should not be forced to decide whether they want to use expensive or restrictive US options or cheaper Chinese solutions. The expensive-and-restricted models in this case are Anthropic’s and OpenAI’s.

Open-weight Chinese models raise security questions

The committee’s focus on a specific concern: open-weight models can be downloaded, modified, and deployed by almost anyone because their core model weights are publicly available. The release says these models lack guardrails against bad actors injecting malicious code or launching attacks on other AI systems.

That portability is how Chinese-origin models have reached American enterprise environments, including through cloud services offered by Amazon, Microsoft, and Google. The committees are examining where the models were trained, whether their capabilities came from unauthorized distillation of leading US systems, and how they spread domestically.

As reported previously by Cryptopolitan, the White House science director Michael Kratsios had accused Moonshot of covertly distilling Anthropic’s Fable to develop Kimi K3 based on the report of Anthropic that had revealed more than 3.4 million conversations of Claude from the company using fake accounts.

DoorDash benchmark also points to US model gains

DoorDash’s own numbers explain the pull. Its AI research lab posted on X that both Kimi K2.6 and Anthropic’s Fable 5 outperformed the Anthropic models the company had been running, and cost less doing it. The committees cited the Kimi half. The same benchmark showed an American frontier model beating its predecessor on price and capability, which is the outcome the letter’s own supply argument is asking for.

Washington remains split on open AI rules

On July 24, Nvidia chief executive Jensen Huang used his first post on X to share an open letter signed by Nvidia, Microsoft, Dell, and Palantir arguing that open-weight models strengthen safety, cybersecurity, and sovereignty.

The Little Tech Association, roughly 200 startups including Y Combinator and Proton, and investors Bill Gurley and Ben Horowitz have argued that restricting the category would mostly entrench Anthropic and OpenAI.

The Trump administration has separately weighed barring US companies from running advanced Chinese open-source models, and Treasury Secretary Scott Bessent said on July 21 that the US could sanction China over AI model theft.

The smartest crypto minds already read our newsletter. Want in? Join them.