OpenAI Labels DeepSeek 'State-Controlled,' Urges Ban on 'PRC-Produced' Models

In a recent policy proposal, OpenAI has labeled the Chinese AI lab DeepSeek as "state-subsidized" and "state-controlled," suggesting that the U.S. government should contemplate banning models developed by DeepSeek and similar operations supported by the People's Republic of China (PRC).

This proposal, submitted as part of the Trump administration's "AI Action Plan" initiative, argues that DeepSeek's models, such as its R1 "reasoning" model, pose security risks due to Chinese legal requirements that mandate compliance with government requests for user data. OpenAI recommends banning "PRC-produced" models in countries classified as "Tier 1" under the Biden administration's export rules to mitigate privacy and security concerns, including the risk of intellectual property theft.

It's not clear whether OpenAI's use of "models" includes DeepSeek's API, its open models, or both. Interestingly, DeepSeek's open models don't have built-in mechanisms for the Chinese government to access user data; instead, they are hosted by companies like Microsoft, Perplexity, and Amazon.

OpenAI has previously accused DeepSeek, which gained recognition earlier this year, of extracting knowledge from OpenAI's models in violation of its terms of service. The new claims—that DeepSeek is backed and controlled by the PRC—mark a significant escalation in OpenAI's ongoing dispute with the Chinese lab.

There's no direct evidence linking DeepSeek, a spin-off from the quantitative hedge fund High-Flyer, to the Chinese government. However, the PRC has shown a growing interest in DeepSeek recently, highlighted by a meeting between DeepSeek founder Liang Wenfeng and Chinese leader Xi Jinping a few weeks ago.

Updated 3/15 8:38 p.m. Pacific: Days after this story hit the news, OpenAI spokesperson Liz Bourgeois provided the following clarification:

“We're not pushing for restrictions on individuals using models like DeepSeek. Our proposal focuses on modifying U.S. export rules to enable more countries to access U.S. computing resources, provided their datacenters don't use PRC technology that could pose security risks. This isn't about limiting chip access based on fears they'll be redirected to the PRC. Our aim is to increase compute and AI access for more countries and people.”

OpenAI's own AI-powered deep research tool describes this statement as "equivocal, using deflective and softening language that somewhat contradicts the more assertive position in the original submission."