OpenAI, the company behind ChatGPT, says it has proof that the Chinese start-up DeepSeek used its technology to create a competing artificial intelligence model — fueling concerns about intellectual property theft in the fast-growing industry.
OpenAI believes DeepSeek, which was founded by math whiz Liang Wenfeng, used a process called “distillation,” which helps make smaller AI models perform better by learning from larger ones.
While this is common in AI development, OpenAI says DeepSeek may have broken its rules by using the technique to create its own AI system.
“The problem is when someone takes our technology and uses it to build their own product,” a source close to OpenAI told Financial Times on Wednesday.
Americans have long been concerned about intellectual property theft from China, particularly in industries like artificial intelligence, biotechnology and advanced manufacturing.
This fear stems from repeated allegations that Chinese companies and state-backed entities engage in cyber espionage, trade secret theft and forced technology transfers to gain a competitive edge in the global market.
The US Navy has officially banned its members from using DeepSeek out of fear the Chinese government could exploit sensitive data, according to a report.
OpenAI has not shared details about the evidence it has found to support the claim, according to FT.
However, its terms of service clearly state that users are not allowed to copy its technology or use its output to build competing AI models.
DeepSeek’s recent AI model, called R1, has been performing as well as leading AI models developed in the US despite being built on a much smaller budget.
This has surprised many in the tech industry and raised questions about how DeepSeek achieved such results.
DeepSeek’s emergence prompted a massive sell-off on Wall Street on Monday.
Nvidia, a major supplier of AI hardware, saw its stock price drop by 17%, losing $589 billion in value.
Investors feared that companies like DeepSeek might no longer need expensive AI hardware.
However, Nvidia’s stock bounced back by 9% on Tuesday as tech stocks recovered.
Last year, OpenAI and its partner Microsoft investigated and blocked accounts that they suspected belonged to DeepSeek.
These accounts had been using OpenAI’s tools in ways that might have violated its rules, sources told FT.
The investigations into the suspicious accounts were first reported by Bloomberg News.
The Post has sought comment from OpenAI and DeepSeek.
Tech leaders in recent days raised the possibility that DeepSeek benefited from intellectual property theft.
David Sacks, an advisor on AI and cryptocurrency to President Trump, suggested that DeepSeek may have stolen OpenAI’s technology.
“There is a technique in AI where one model learns from another by copying its knowledge. There is strong evidence that DeepSeek did this with OpenAI’s models,” Sacks told Fox News.
However, he did not provide concrete proof.
DeepSeek claims it used only 2,048 Nvidia H800 graphics cards and spent $5.6 million to train its AI model, which has 671 billion parameters.
This is far less than what companies like OpenAI and Google spend on similar models.
Some AI experts say DeepSeek’s model seems to have been trained using OpenAI’s GPT-4, which would violate OpenAI’s terms of service.
Experts say it is common for AI start-ups in China and the US to use results from established AI models to improve their own systems.
“Many start-ups and researchers use responses from commercial AI models like ChatGPT to train their own models,” Ritwik Gupta, a PhD candidate at the University of California, Berkeley, told FT.
“It saves time and money because they don’t have to gather human feedback themselves.”
The challenge for OpenAI is how to protect its technology while still making AI accessible.
The company says it actively defends its intellectual property.
“We know that companies, including those in China, are always trying to copy our AI models,” OpenAI said in a statement to FT.
OpenAI also noted that it works closely with the US government to prevent adversaries from stealing advanced AI technology.
Ironically, OpenAI itself is facing lawsuits for allegedly using copyrighted material without permission.
The New York Times and several publishers have sued OpenAI, accusing it of training ChatGPT on their work without proper authorization.
In May of last year, OpenAI and News Corp struck a multi-year deal granting OpenAI access to current and archived content from News Corp’s major publications, including The Post, the Wall Street Journal, The Times of London and The Australian.
OpenAI has established similar content partnerships with other publishers, including Condé Nast, Le Monde and Prisa Media.
