In a new policy proposal, OpenAI describes Chinese AI lab DeepSeek as “state-subsidized” and “state-controlled,” and recommends that the U.S. government consider banning models from the outfit and similar People’s Republic of China (PRC)-supported operations.
The proposal, a submission for the Trump administration’s “AI Action Plan” initiative, claims that DeepSeek’s models, including its R1 “reasoning” model, are insecure because DeepSeek faces requirements under Chinese law to comply with demands for user data. Banning the use of “PRC-produced” models in all countries considered “Tier 1” under the Biden administration’s export rules would prevent privacy and “security risks,” OpenAI says, including the “risk of IP theft.”
It’s unclear whether OpenAI’s references to “models” are meant to refer to DeepSeek’s API, the lab’s open models, or both. DeepSeek’s open models don’t contain mechanisms that would allow the Chinese government to siphon user data; companies including Microsoft, Perplexity, and Amazon host them on their infrastructure.
OpenAI has previously accused DeepSeek, which rose to prominence earlier this year, of “distilling” knowledge from OpenAI’s models against its terms of service. But OpenAI’s new allegations — that DeepSeek is supported by the PRC and under its command — are an escalation of the company’s campaign against the Chinese lab.
There isn’t a clear link between the Chinese government and DeepSeek, a spin-off from a quantitative hedge fund called High-Flyer. However, the PRC has taken an increased interest in DeepSeek in recent months. Several weeks ago, DeepSeek founder Liang Wenfeng met with Chinese leader Xi Jinping.
Updated 3/15 8:38 p.m. Pacific: Days after this story was published, OpenAI spokesperson Liz Bourgeois sent the following statement:
“We’re not advocating for restrictions on people using models like DeepSeek. What we’re proposing are changes to U.S. export rules that would allow additional countries to access U.S. compute on the condition that their datacenters don’t rely on PRC technology that present[s] security risks — instead of restricting their access to chips based on the assumption that they will divert technology to the PRC. The goal is more compute and more AI for more countries and more people.”
Disrupt 2026: The tech ecosystem, all in one room
Your next round. Your next hire. Your next breakout opportunity. Find it at TechCrunch Disrupt 2026, where 10,000+ founders, investors, and tech leaders gather for three days of 250+ tactical sessions, powerful introductions, and market-defining innovation. Register now to save up to $400.
Save up to $300 or 30% to TechCrunch Founder Summit
1,000+ founders and investors come together at TechCrunch Founder Summit 2026 for a full day focused on growth, execution, and real-world scaling. Learn from founders and investors who have shaped the industry. Connect with peers navigating similar growth stages. Walk away with tactics you can apply immediately
Offer ends March 13.
OpenAI’s own AI-powered deep research tool characterizes the above statement as “equivocal, employing deflective and softening language that partially contradicts the stronger stance documented in [the company’s] original submission.”
