Microsoft's security researchers have spotted some suspicious activity. They noticed large-scale data extraction from OpenAI's API in late 2024, according to a report from Bloomberg.
Microsoft and OpenAI are now looking into whether a group linked to the Chinese AI startup DeepSeek improperly accessed data from OpenAI's technology. This is a serious concern.
In late 2024, Microsoft flagged unusual data extraction through OpenAI’s application programming interface (API). They alerted OpenAI about this activity. It might violate OpenAI's terms of service, or the group may have bypassed restrictions on data collection.
On January 20, DeepSeek launched its new AI model, R-1. This model reportedly competes with ChatGPT and does so at a much lower cost. The announcement caused a drop in tech and AI stocks, leading to billions lost in the U.S. market.
David Sacks, the White House's crypto and AI czar, shared some insights on January 28. He mentioned evidence that DeepSeek used OpenAI's model outputs to train their AI. This process is called distillation, where one AI model learns from another's outputs.
“There’s substantial evidence that what DeepSeek did is distill knowledge from OpenAI models. I don’t think OpenAI is very happy about this,” Sacks stated.
OpenAI has acknowledged concerns about Chinese companies trying to distill U.S. AI models, but they didn’t specifically comment on DeepSeek.
“We know that companies based in China and elsewhere are constantly trying to distill the models of leading U.S. AI companies,” an OpenAI spokesperson told Bloomberg.
This investigation focuses on possible violations of OpenAI's terms of service and the potential circumvention of data access limits. It highlights the growing tensions in the competition for AI technology between the U.S. and China.
Additionally, CNBC reported on January 28 that the U.S. Navy has banned its personnel from using DeepSeek. They are concerned that the Chinese government could exploit sensitive data.
The Navy sent a warning email to its members on January 24. They stated that the AI model should not be used “in any capacity” due to potential security and ethical concerns regarding its origin and use.