LOADING...
OpenAI labels DeepSeek as 'China-controlled,' urges ban on such models
OpenAI has submitted a new policy proposal

OpenAI labels DeepSeek as 'China-controlled,' urges ban on such models

Mar 14, 2025
11:58 am

What's the story

OpenAI has called out Chinese AI start-up DeepSeek as "state-subsidized" and "state-controlled," in a new policy proposal. The organization is calling on the US government to consider banning models from DeepSeek and other People's Republic of China (PRC)-backed operations. The proposal, submitted for Trump Administration's "AI Action Plan" initiative, raises concerns over user data security under Chinese law.

Risk assessment

DeepSeek's models seen as security threat

OpenAI's proposal notably emphasizes DeepSeek's R1 "reasoning" model, calling it insecure as it is required to comply with user data requests under Chinese law. The company contends that banning all "PRC-produced" models in "Tier 1" countries under the Biden Administration's export rules could reduce privacy and security risks, including intellectual property theft.

Prior allegations

OpenAI's previous accusations against DeepSeek

Previously, OpenAI had accused DeepSeek of "distilling" knowledge from its models, violating the terms of service. The latest allegations mark an escalation in OpenAI's campaign against the Chinese AI lab. However, there is no clear evidence linking DeepSeek to the Chinese government, despite increasing interest from PRC in recent months.

Data concerns

DeepSeek's open models and their security

It remains unclear if OpenAI's references to "models" are limited to DeepSeek's API or also include the lab's open models. These open models, hosted on Microsoft, Amazon, and Perplexity infrastructure, do not have mechanisms that would allow the Chinese government to access user data. This raises questions about the security of these models in relation to OpenAI's concerns.

Information

Permit AI training on copyrighted content

Separately, OpenAI has urged the Trump administration to ease regulations, advocating policies that permit AI companies to use copyrighted content for training. The goal is to boost AI development and enhance competition against China in the rapidly evolving AI race.