Chinese startup DeepSeek's recently launched open-source reasoning model R1 has quickly ignited the entire AI technology field. DeepSeek's "low-cost, high-performance" features have impacted the American tech giants that previously dominated the AI landscape, with NVIDIA, OpenAI, and others seeing their stock prices plummet. Even U.S. President Trump has referred to DeepSeek as the "wake-up call for the U.S. AI industry."
DeepSeek Suspected of Infringing OpenAI's Intellectual Property Rights?
However, according to a report by the Financial Times published earlier today (29th), OpenAI has revealed that it has found evidence indicating that DeepSeek may have used OpenAI's proprietary models to train its products, potentially infringing on OpenAI's intellectual property rights.
OpenAI further explained that DeepSeek is suspected of using a technique called "distillation," which can help DeepSeek leverage OpenAI's models to improve its performance without having to incur the high costs of training its own models:
Distillation is not uncommon in the industry, and OpenAI has provided legal ways for developers to use this technique. However, the issue arises if you are using it to create a competing product to OpenAI.
Additionally, David Sacks, the AI czar appointed by U.S. President Trump, has also expressed the view that DeepSeek may have infringed on intellectual property rights.
Microsoft Assists OpenAI in Investigating DeepSeek
Furthermore, sources have revealed that Microsoft, another major American tech company, has been working with OpenAI since last fall to investigate accounts suspected to belong to DeepSeek, and they had already discovered that DeepSeek was using OpenAI's API for distillation. Subsequently, OpenAI has blocked DeepSeek's accounts due to violations of its terms of use.
Currently, OpenAI has issued a statement indicating that it will take countermeasures to protect its intellectual property rights:
We are aware that Chinese companies and others are constantly trying to use distillation techniques to steal the models of top AI companies in the U.S.
We will take countermeasures to protect our intellectual property, including carefully deciding which cutting-edge capabilities to include in our released models.
We believe that future cooperation with the U.S. government is crucial, as we need to protect the most powerful AI models and prevent them from being stolen by competitors.
Is OpenAI Having Difficulty Protecting Its Rights?
However, it should be noted that since distillation is a common practice in AI model training, efforts to prohibit such behavior may face various challenges. For example, Ritwik Gupta, an AI Ph.D. from the University of California, stated:
Startups and academic institutions often use commercially available large models like ChatGPT that have been human-aligned for training their own models, allowing them to obtain top-tier model research results at low cost.
I'm not surprised if DeepSeek has adopted this approach, and if they have, it will be very difficult to completely prevent such behavior.
Additionally, the Financial Times further noted that OpenAI itself is facing allegations of copyright infringement, with The New York Times and several prominent authors filing lawsuits accusing it of using news articles and book content to train its AI models without permission.
