OpenAI Accuses DeepSeek of Being 'State-Controlled', Demands Bans
OpenAI Calls for Potential Ban of Chinese AI Lab DeepSeek
A recently unveiled policy proposal from OpenAI characterizes the Chinese AI laboratory, DeepSeek, as being both “state-subsidized” and “state-controlled.” The company suggests the U.S. government contemplate prohibiting the use of models originating from DeepSeek and comparable entities backed by the People’s Republic of China (PRC).
Concerns Regarding Data Security and Intellectual Property
This proposal, submitted as part of the Trump administration’s “AI Action Plan” initiative, asserts that DeepSeek’s models – notably its R1 “reasoning” model – present security vulnerabilities. These vulnerabilities stem from DeepSeek’s legal obligations under Chinese law to fulfill data requests from authorities.
OpenAI argues that a ban on “PRC-produced” models within nations classified as “Tier 1” under the Biden administration’s export regulations would mitigate privacy and security risks. Specifically, the company highlights the potential for IP theft.
Scope of the Proposed Restrictions
The precise definition of “models” within OpenAI’s proposal remains ambiguous. It is unclear whether the restrictions are intended to encompass DeepSeek’s API, its publicly available models, or both.
It's important to note that DeepSeek’s open-source models do not inherently possess mechanisms for the Chinese government to extract user data. These models are currently hosted on the infrastructure of companies like Microsoft, Perplexity, and Amazon.
Escalation of Concerns Regarding DeepSeek
OpenAI has previously leveled accusations against DeepSeek, alleging that the lab engaged in “distilling” knowledge from OpenAI’s models in violation of their terms of service. However, the current claims – asserting PRC support and control over DeepSeek – represent a significant intensification of OpenAI’s opposition to the Chinese lab.
Government Ties and Recent Developments
While a direct connection between the Chinese government and DeepSeek remains unconfirmed, the PRC has demonstrated growing interest in the company in recent months.
Notably, DeepSeek’s founder, Liang Wenfeng, recently met with Chinese leader Xi Jinping, signaling increased attention from the highest levels of the Chinese government.
Clarification from OpenAI
Following the initial publication of this story, OpenAI spokesperson Liz Bourgeois issued the following statement:
“Our intention is not to restrict individuals from utilizing models such as DeepSeek. Instead, we are advocating for modifications to U.S. export regulations. These changes would allow broader access to U.S. computing resources for additional countries, contingent upon their data centers avoiding the use of PRC technology that poses security risks. This approach differs from restricting access to chips based on assumptions of potential diversion to the PRC.”
“The ultimate goal is to increase the availability of computing power and AI capabilities for a wider range of nations and individuals.”
Internal Assessment of OpenAI’s Statement
OpenAI’s own AI-powered research tool has analyzed the spokesperson’s statement, characterizing it as “equivocal.” The tool notes the use of “deflective and softening language” that appears to partially contradict the more assertive position outlined in the company’s original submission.
