In the new policy proposal, Openai describes China’s AI lab Deepseek as “state aid” and “state management,” and recommends that the US government consider banning models from costumes and similar projects supported by the People’s Republic (PRC).
The proposal, a proposal to the Trump administration’s “AI Action Plan” initiative, claims that Deepseek’s model, including the R1’s “inference” model, is unsettling as it faces requirements under Chinese law to comply with user data requirements. Prohibiting the use of the “PRC production” model in all countries considered “Tier 1” under Biden Administration’s export regulations will prevent privacy and “security risks,” according to Openai, which includes the “risk of IP theft.”
It is unclear whether the reference to Openai’s “model” is intended to refer to DeepSeek’s API, the lab’s open model, or both. Deepseek’s open model does not include a mechanism by which the Chinese government can siphon user data. Companies like Microsoft, Perplexity, and Amazon host them on their infrastructure.
Openai previously denounced Deepseek, who “distilled” knowledge from Openai’s model earlier this year, and “distilled” its knowledge on its terms and conditions. But Openai’s new allegation – Deepseek is supported by the PRC, under its command, an escalation of the company’s campaign against Chinese labs.
There is no clear link between the Chinese government and Deepseek. This is a spinoff from a quantitative hedge fund called High-Flyer. However, PRC has been gaining interest in DeepSeek in recent months. A few weeks ago, Deepseek founder Liang Wenfeng met with Chinese leader Xi Jinping.