OpenAI says Chinese companies are trying to use US models to train AI

ChatGPT creator OpenAI has evidence that the Chinese artificial intelligence (AI) app DeepSeek used the US-based company’s technology to train its budget model, according to a report in the Financial Times.

An OpenAI spokesperson told Euronews Next that Chinese companies and others were "constantly trying to distil the models of leading US AI companies," in response to a request for comment on the report.

Distillation is the practice of leveraging a larger AI model to transfer information to a smaller one.

An OpenAI spokesperson added that the company works to protect its intellectual property and that it is critically important to protect "the most capable models from efforts by adversaries and competitors to take US technology".

According to OpenAI, distillation is prohibited under their Terms of Service and they have measures in place to prevent and detect such actions.

The company also said they have investigated previous attempts to distil their models and are working with Microsoft, which has invested billions in OpenAI, to identify and respond to actors trying to do so.

'Could be one explanation'

Anthony Cohn, a professor of automated reasoning at the University of Leeds in the UK, said in an email to Euronews Next that "without the kind of detailed data and knowledge that only OpenAI itself has, or being privy to exactly how a new [large language model] has been trained, an outsider cannot be sure whether the OpenAI claim is true or not".

"This could be one explanation of the quality of the DeepSeek model, but not the only one," he added.

"It might also explain why the quoted training costs are so low, but again without confidential company information no one can be sure".

He also pointed out that AI models are often trained with data taken from the web "which many argue infringes copyright restrictions, and so the question of whether anyone is completely 'clean' is unclear".

‘Substantial evidence’ the Chinese distilled OpenAI models

The White House’s AI czar David Sacks, meanwhile, told US broadcaster Fox News in an interview on Tuesday that it was “possible” that there was intellectual property theft.

Sacks explained that distillation is when "the student model asks the parent model a lot of questions” with the AI asking millions of questions to “suck the knowledge out of the parent model".

"There’s substantial evidence that what DeepSeek did here is that they distilled the knowledge out of OpenAI’s models," Sacks said.

The warnings come after DeepSeek disrupted global markets this week with the release of its latest model which it said performs on par with OpenAI.

In reaction to the new model’s release, OpenAI CEO Sam Altman wrote in a post on X that it was an "impressive model" particularly regarding "what they’re able to deliver for the price".

He added that it was "invigorating to have a new competitor".