OpenAI, the creator of ChatGPT, claims to have evidence that Chinese artificial intelligence startup DeepSeek used OpenAI‘s proprietary models to train its competing AI system.
The San Francisco-based company discovered what it believes to be signs of “distillation” by DeepSeek, a technique where developers use outputs from larger AI models to improve the performance of smaller ones, according to the Financial Times.
Microsoft’s security researchers detected suspicious activity last fall, when individuals believed to be connected to DeepSeek extracted large amounts of data through OpenAI’s application programming interface (API).
While distillation is a common practice in AI development, OpenAI maintains that DeepSeek’s alleged use violated its terms of service, which explicitly prohibit using outputs to develop competing models. The company has already taken action by blocking suspected DeepSeek accounts that were accessing OpenAI’s API last year, the Financial Times reported.
DeepSeek recently gained attention for developing its R1 reasoning model at a fraction of the cost typically associated with training large AI models. The company claims to have spent just $5.6 million using 2,048 Nvidia H800 graphics cards to train its model with 671 billion parameters.
David Sacks, the White House AI advisor, addressed the situation on Fox News, suggesting there was “substantial evidence” that DeepSeek had extracted knowledge from OpenAI’s models, though he did not provide specific details.
OpenAI expressed broader concerns about Chinese companies’ attempts to replicate US AI technology, stating they are “constantly trying to distil the models of leading US AI companies.” The company emphasized its commitment to working with the US government to protect advanced AI capabilities from competitors and adversaries.