OpenAI says it has evidence China’s DeepSeek used its model to train competitor

Stay informed with free updates

OpenAI says it has found evidence that Chinese artificial intelligence start-up DeepSeek used the US company’s proprietary models to train its own open-source competitor, as concerns grow over a potential breach of intellectual property.

The San Francisco-based ChatGPT maker told the Financial Times it had seen some evidence of “distillation”, which it suspects to be from DeepSeek.

The technique is used by developers to obtain better performance on smaller models by using outputs from larger, more capable ones, allowing them to achieve similar results on specific tasks at a much lower cost.

Distillation is a common practice in the industry but the concern was that DeepSeek may be doing it to build its own rival model, which is a breach of OpenAI’s terms of service.

“The issue is when you [take it out of the platform and] are doing it to create your own model for your own purposes,” said one person close to OpenAI.

OpenAI declined to comment further or provide details of its evidence. Its terms of service state users cannot “copy” any of its services or “use output to develop models that compete with OpenAI”.

DeepSeek’s release of its R1 reasoning model has surprised markets, as well as investors and technology companies in Silicon Valley. Its built-on-a-shoestring models have attained high rankings and comparable results to leading US models.

Shares in Nvidia fell 17 per cent on Monday, wiping $589bn off its market value, on fears that big investments in its expensive AI hardware might not be needed. They recovered by 9 per cent on Tuesday, along with other tech stocks.

OpenAI and its partner Microsoft investigated accounts believed to be DeepSeek’s last year that were using OpenAI’s application programming interface (API) and blocked their access on suspicion of distillation that violated the terms of service, another person with direct knowledge said. These investigations were first reported by Bloomberg.

Microsoft declined to comment and OpenAI did not immediately respond to a request for comment on this detail. DeepSeek did not respond to a request for comment made during the lunar new year holiday.

Earlier, President Donald Trump’s AI and crypto tsar David Sacks said “it is possible” that IP theft had occurred.

“There’s a technique in AI called distillation . . . when one model learns from another model [and] kind of sucks the knowledge out of the parent model,” Sacks told Fox News on Tuesday.

“And there’s substantial evidence that what DeepSeek did here is they distilled the knowledge out of OpenAI models, and I don’t think OpenAI is very happy about this,” Sacks added, although he did not provide evidence.

DeepSeek said it used just 2,048 Nvidia H800 graphics cards and spent $5.6mn to train its V3 model with 671bn parameters, a fraction of what OpenAI and Google spent to train comparably sized models. Some experts said the model generated responses that indicated it had been trained on outputs from OpenAI’s GPT-4, which would violate its terms of service. 

Industry insiders say that it is common practice for AI labs in China and the US to use outputs from companies such as OpenAI, which have invested in hiring people to teach their models how to produce responses that sound more human. This is expensive and labour-intensive, and smaller players often piggyback off this work, say the insiders.

“It is a very common practice for start-ups and academics to use outputs from human-aligned commercial LLMs, like ChatGPT, to train another model,” said Ritwik Gupta, a PhD candidate in AI at the University of California, Berkeley.

“That means you get this human feedback step for free. It is not surprising to me that DeepSeek supposedly would be doing the same. If they were, stopping this practice precisely may be difficult,” he added.

The practice highlights the difficulty for companies keen to protect their technical edge. “We know [China]-based companies — and others — are constantly trying to distil the models of leading US AI companies,” OpenAI said in its latest statement.

It added: “We engage in countermeasures to protect our IP, including a careful process for which frontier capabilities to include in released models, and believe . . . it is critically important that we are working closely with the US government to best protect the most capable models from efforts by adversaries and competitors to take US technology.”

OpenAI is battling allegations of its own copyright infringement from newspapers and content creators, including lawsuits from The New York Times and prominent authors, who accuse the company of training its models on their articles and books without permission.

Source link

Visited 1 times, 1 visit(s) today

Related Article

Headshot of Brandon J. Weichert

Does China Have the Most Advanced Submarine Detection Technology?

China is on a mission to become the world’s dominant superpower by 2049. That goal is driving a desire to build a world-class navy that can challenge the United States. To make it even stronger, according to Stephen Chen of the South China Morning Post, scientists from Shanghai Jiao Tong University have created a new

Shoppers purchase beef at a supermarket in southeastern China. The country’s meat consumption has risen rapidly over the past decade. Photo: Getty Images

China’s groundbreaking algorithm, cyberattack on DeepSeek: SCMP daily highlights

Catch up on some of SCMP’s biggest China and economy stories of the day. If you would like to see more of our reporting, please consider subscribing. 1. Chinese algorithm boosts Nvidia GPU performance 800-fold in science computing A high-performance algorithm that could solve complicated material design problems on consumer GPUs has been developed by

IT minister Ashwini Vaishnaw on privacy issues with China's DeepSeek: They can be addressed by ...

IT minister Ashwini Vaishnaw on privacy issues with China’s DeepSeek: They can be addressed by …

IT Minister Ashwini Vaishnaw has underscored India’s expanding artificial intelligence (AI) capabilities, emphasizing the country’s access to high-end computing resources and the feasibility of hosting open-source AI models like DeepSeek while addressing privacy concerns.“It is an open-source model. Like LLaMA (LLM by Meta), which is open source, this too can be hosted on Indian servers.Data

Deepseek IA china

DeepSeek is no game: The dangers of China’s new AI | Technology

Welcome greeting from Chinese artificial intelligence app Deepseek.Dado Ruvic (REUTERS) In just one week, DeepSeek, China’s new artificial intelligence app, has surpassed one million downloads (more than half in just two days) and it is expected to multiply that figure, according to AppBrain based on data from QR Code Generator. But this technological frenzy has

French Aircraft Carrier Visits Indonesia

NATO Aircraft Carrier Watched by China Reaches Pacific Waters

A nuclear-powered aircraft carrier sent by France, a NATO member, has reached the edge of the Western Pacific Ocean, where its deployment is being closely monitored by China. The Chinese Defense Ministry did not immediately respond to Newsweek‘s request for comment. Newsweek has contacted NATO‘s press office for comment via its website. Why It Matters

China’s DeepSeek model is a major advance in AI technology

China’s DeepSeek model is a major advance in AI technology

The logo for the app DeepSeek [AP Photo/Jon Elswick] Last week, DeepSeek, a startup company based in Hongzhou, China, released its newest artificial intelligence model, DeepSeek R1. Within days, the chatbot became the most-downloaded app in Apple’s App Store. DeepSeek’s performance meets or exceeds that of state-of-the-art AI models from American companies such as Meta

Su Min’s divorce certificate has come through. Photo: Weibo/微博@50歲阿姨自駕遊

China’s ‘road-tripping auntie’ vows to keep going after divorce comes through

“Auntie” Su Min, a Chinese grandmother who has inspired countless women after leaving her abusive husband to go on a solo road trip in 2020, announced her divorce on Monday after nearly four decades of marriage. Vowing to “stay on the road”, the 60-year-old made the announcement on her Weibo social media account where she

Latest and Breaking News on NDTV

What China’s DeepSeek AI Says

New Delhi: Chinese artificial intelligence (AI) startup DeepSeek claims to have developed an AI assistant with performance comparable to leading Western models like OpenAI’s ChatGPT and Google’s Gemini but at a fraction of the cost. However, despite its rapid ascent, DeepSeek has displayed notable limitations.  Like other Chinese AI models, it remains constrained by government

Roger Dassen, CFO of ASML, speaks at the announcement of fourth quarter results in Veldhoven, Netherlands, January 29 2025. Photo: Reuters

China years behind ASML’s chipmaking machines, CFO Dassen says

It will take China years to catch up with ASML Holding NV’s chipmaking technology, according to a senior executive at the Dutch firm. “Never say never,” Chief Financial Officer Roger Dassen said in an interview at company’s headquarters in Veldhoven, the Netherlands, adding that ASML is far ahead of Chinese competition. “Our technology is very

230105_navy_digital_gfx

To be ready for China in 2027, think new software, not ships: Navy tech officer

The Navy is working to digitize its infrastructure and operations. (Graphic by Breaking Defense, original images via Getty and DVIDS) WEST 2025 — The way the US will beat the People’s Republic of China (PRC) in a near-term Indo-Pacific conflict won’t be with new ships, but with a focus on cutting-edge software and artificial intelligence

A.I. leaders say China may have copied U.S. technology

A.I. leaders say China may have copied U.S. technology

IE 11 is not supported. For an optimal experience visit our site on another browser. UP NEXT Report: Scores in reading decline further among 4th and 8th graders 01:53 Three more Israeli hostages to be released by Hamas 00:49 Pilot survives fiery crash of F-35 fighter plane 01:29 Kennedy grilled by Senate Democrats on vaccines

China's shocking DeepSeek AI pops US Big Tech monopoly bubble

China’s shocking DeepSeek AI pops US Big Tech monopoly bubble

China is making enormous progress in the development of artificial intelligence technology, and it has set off a political and economic earthquake in the West. The stocks of US Big Tech corporations crashed on January 27, losing hundreds of billions of dollars in market capitalization over the span of just a few hours, on the

Two dancers in costume hold each other's  arms as one stands and one spins, his feet in the air

China’s biggest holiday: Lunar New Year and how it’s celebrated

Among China’s traditional holidays and celebrations, none ranks higher in importance than the Lunar New Year (農曆新年). Also known as the Spring Festival (春節), or simply Chinese New Year, it marks the beginning of the year according to the traditional lunar calendar. The Lunar New Year usually starts sometime between late January and mid-February. This

Nvidia stock extends losses after report of potential additional curbs on China sales

Nvidia (NVDA) stock sank 6% Wednesday after Bloomberg reported that Trump administration officials are “exploring additional curbs” on Nvidia’s chip sales to China. Bloomberg’s report added that the talks are in the “very early stages.” The new restrictions would likely cover Nvidia’s H20 chips, which is a scaled-down product the chipmaker offers to meet existing

Authorities process a man who has his hands on the side of a van

Chinese migrants, Cuban smugglers pour into South Florida from Bahamas by boat

Amid a nationwide crackdown on known criminal aliens in the first days of President Donald Trump’s second arrival at the White House, police and federal agents have stumbled upon dozens of Chinese migrants in Coral Gables, Florida, and arrested at least four suspected smugglers from Cuba and another from Haiti. Now local authorities are hoping

IndyTech

How DeepSeek users are forcing the AI to reveal the truth about Chinese executions

Sign up to our free weekly IndyTech newsletter delivered straight to your inbox Sign up to our free IndyTech newsletter Sign up to our free IndyTech newsletter There are certain things DeepSeek doesn’t want you to know. 1,156 things, to be precise. The Chinese startup’s latest AI chatbot went viral this week after its performance

0
Would love your thoughts, please comment.x
()
x