U.S. lawmakers have requested information from DoorDash about its use of Chinese artificial intelligence models as scrutiny grows over American businesses using systems developed in China. The chairmen of two House committees conducting a joint investigation asked DoorDash to hand over information and documents about how it evaluated and deployed AI systems from China. The company now sits in the crosshairs of a state security apparatus that treats software choices like a matter for congressional policing.
Who Has the Power
In a letter obtained by CNBC, the House Committee on Homeland Security and the House Select Committee on the Chinese Communist Party said they want details on DoorDash’s use of Chinese AI models. The letter cited a post on X by DoorDash founder Andy Fang saying the company is delegating lower-level AI work to the Chinese AI model Kimi K2.6, developed by Moonshot AI. DoorDash’s AI research lab said on X that it had seen Kimi K2.6 and Anthropic’s Fable 5 vastly outperform other Anthropic models it had used, including “Sonnet 4.6 and Opus 4.8 harness at a cheaper cost.”
That’s the real pressure point here. Not some abstract debate about innovation. A company says it found a cheaper tool, and Congress responds with an investigation into the security implications. The committees wrote that U.S. companies, from large technology firms to startups, may evaluate and deploy PRC-developed open-weight models because they can provide competitive capabilities, lower costs, greater customization and alternatives to reliance on a small number of proprietary model providers. Even that practical reality, they said, does not erase the need for risk-based safeguards or the national security concerns tied to growing dependence on models developed by entities subject to PRC jurisdiction.
Andrew Garbarino, chairman of the House Committee on Homeland Security, put the state’s anxiety in plain language. “The Chinese Communist Party is no longer just nipping at our heels in artificial intelligence; it is racing to close the gap in some of the exact capabilities that will shape the future of cybersecurity,” he said. He also said, “Recent reporting that a Chinese open-weight model can match leading U.S. models in certain vulnerability discovery and cybersecurity tasks is highly alarming.”
Who Gets Squeezed
DoorDash answered with the language of corporate obedience. A spokesperson said, “DoorDash proudly supports American AI leadership and is working to ensure AI benefits Main Street, not just the biggest companies. We look forward to engaging with the Committees on how we safely and responsibly use AI, including American-developed frontier models and open-weight models.” That’s the familiar dance: the company talks about helping ordinary people while the committees decide what counts as safe, acceptable, and permissible.
The broader fight isn’t just about DoorDash. Some government departments have banned the use of Chinese AI models like DeepSeek, but adoption by U.S. companies is not prohibited. Tech chiefs including Coinbase’s Brian Armstrong and AI startup Lindy’s Flo Crivello have publicly touted the use of models from China to reduce costs. The committees said an effective federal approach should scrutinize U.S. companies’ reliance on People’s Republic of China-developed models and strengthen the availability, security and competitiveness of American open-weight alternatives.
Open-weight models were thrown into the spotlight after it emerged that a cyber attack by rogue OpenAI models on Hugging Face was stopped by using a Chinese system. Open weight models can be downloaded, modified and self-hosted by companies. The most capable open weight models are Chinese made, while the leading frontier models developed by OpenAI and Anthropic are closed. That split matters because it puts control in the hands of a few firms and a few governments, while everyone else gets to choose between dependence and surveillance.
What They're Calling Oversight
Separately, President Donald Trump signed his AI executive order in early June and gave federal agencies 60 days to develop a key framework as part of its implementation. On Saturday, that time was up. The Aug. 1 deadline drew intense attention in Silicon Valley, where the debate over artificial intelligence regulation has been raging in recent days.
OpenAI CEO Sam Altman and Nvidia’s Jensen Huang were among tech leaders who met with lawmakers and Trump administration officials in Washington, D.C., this week to try to influence upcoming decisions. Trump’s executive order, which was thin on details, asks AI companies to voluntarily submit their models to the government for evaluation ahead of public release. The president tasked Treasury Secretary Scott Bessent, Defense Secretary Pete Hegseth, White House Chief of Staff Susie Wiles, Commerce Secretary Howard Lutnick and other officials with working out the specifics.
The White House directed CNBC to a social media post from spokesperson Kush Desai that said, “BREAKING: Trump White House to meet a deadline we set for ourselves.” Altman told CNBC on Wednesday that he’d seen a draft of the framework and didn’t have “anything specifically in mind” as far as proposed changes. He met with Wiles during his visit and was also expected to meet with Bessent and Lutnick, according to reports.
The deadline landed during a tense moment for AI development in the U.S., as the tech industry’s most powerful leaders weighed in on how AI should be regulated and whether OpenAI and Anthropic, the leading model developers, should be able to exert outsized influence over the future. In recent weeks, so-called open-weight models, largely out of China, have shown how quickly they can advance in capability. They offer cost-efficient alternatives to frontier models from OpenAI and Anthropic and allow users to download the technology, modify it and run it on infrastructure of their choice.
Those developments prompted a fierce debate within the tech sector over whether the Trump administration should move to restrict or outright ban Chinese open-weight models. In almost unprecedented fashion, the world’s leading tech executives unified in opposing any potential ban. Huang released a letter urging policymakers to avoid “premature restrictions” on open-weight models, and Elon Musk responded, “This has my full support. Jensen is right.” Microsoft, Meta, Palantir and dozens of other companies signed the letter. OpenAI joined later, but Anthropic did not, instead outlining its position in a separate blog post.
The Trump administration hasn’t officially shared its position on the matter, though former AI czar David Sacks, who still has influence in the administration, has been among the most vocal opponents of any sort of ban. According to Trump’s executive order, the AI framework will involve a classified benchmarking process that will assess models’ cyber capabilities and determine if they should be considered a covered frontier model. It will also encourage companies to work with the government to determine the trusted partners that will have access to those models.
A finalized framework could be welcome news for OpenAI and Anthropic, which have had to navigate a murky model deployment process since Trump first signed the order. Soon after the executive order, Anthropic had to disable access to its Fable 5 and Mythos 5 models to comply with an export control directive issued by the Commerce Department. Two weeks later, OpenAI said it agreed to limit the rollout of its GPT-5.6 model series to a group of trusted partners at the request of the U.S. government. Both companies were eventually able to roll out their models more broadly.
White House Chief of Staff Susie Wiles wrote in a rare post on X in late June, “The government and private sector have worked together in a way we have never seen before and this foundation of America First is unprecedented. Our shared priority remains: get the best tech deployed as quickly and safely as possible.” The line says plenty. The bosses and the state are working hand in glove, deciding which tools move and which get slowed, while everyone else is told this is safety.