ChatGPT Maker Suspects China’s Dirt Cheap DeepSeek AI Models Were Built Using OpenAI Data — and the Irony Is Not Lost on the Internet

OpenAI suspects that China's DeepSeek AI models, significantly cheaper than Western alternatives like ChatGPT, may have been trained using OpenAI's data. This revelation, coupled with DeepSeek's rapid rise in popularity, triggered a sharp decline in the stock prices of major AI companies, with Nvidia experiencing the most substantial losses in Wall Street history.

DeepSeek's R1 model, based on the open-source DeepSeek-V3, boasts significantly lower training costs (estimated at $6 million) and computational requirements compared to Western counterparts. While this claim is disputed by some, it has fueled investor concerns regarding the massive investments being made by American tech companies in AI. DeepSeek's success has raised questions about the value of these investments.

OpenAI and Microsoft are now investigating whether DeepSeek violated OpenAI's terms of service by employing a technique called "distillation" – extracting data from larger models to train smaller ones – using OpenAI's API. OpenAI acknowledges that Chinese companies are actively attempting to replicate leading U.S. AI models, and is taking countermeasures to protect its intellectual property. They are also advocating for closer collaboration with the U.S. government to safeguard advanced AI models.

David Sacks, President Trump's AI czar, supports OpenAI's concerns, suggesting that DeepSeek's actions constitute a breach of OpenAI's intellectual property rights. He anticipates further measures from leading AI companies to prevent similar incidents.

The situation is further complicated by the irony that OpenAI itself has faced accusations of using copyrighted material without permission in the development of ChatGPT. This hypocrisy has been highlighted by numerous commentators, given OpenAI's previous statements that creating AI models like ChatGPT without copyrighted material is impossible. OpenAI's stance on this issue is further underscored by its submission to the UK's House of Lords and its ongoing legal battles, including a lawsuit from the New York Times and a previous suit by 17 authors. The legal landscape surrounding the use of copyrighted material in training AI models remains highly contested. A 2018 U.S. Copyright Office finding that AI-generated art cannot be copyrighted adds another layer of complexity to the debate.