OpenAI Claims DeepSeek Used Its Proprietary Models To Train Rival System
OpenAI has accused Chinese artificial intelligence start-up DeepSeek of using its proprietary models to develop an open-source competitor, raising concerns about potential intellectual property violations. The US-based company told the Financial Times that it had found evidence of “distillation”—a technique that allows developers to enhance smaller models by using outputs from larger, more capable ones. While distillation is a common industry practice, OpenAI suspects that DeepSeek may have used this method in breach of its terms of service.
An individual close to OpenAI explained that the concern arises when the technique is used outside of the platform to develop competing models. OpenAI’s terms prohibit users from copying its services or using its outputs to build rival models. The company declined to provide further details on its findings.
The allegations come amid growing interest in DeepSeek, which has emerged as an unexpected force in AI development. The start-up’s recently launched R1 reasoning model has drawn significant attention for achieving performance levels comparable to major US-based AI models, despite being built with considerably fewer resources. Reports suggest DeepSeek developed its model using just $6 million worth of computing power—far less than the billions invested by industry leaders like OpenAI and Google.
The rise of DeepSeek has unsettled the tech market. On Monday, Nvidia shares fell by 17 per cent, wiping $589 billion off its market value, as investors feared that expensive AI hardware might become less critical. However, Nvidia’s stock rebounded by 9 per cent the following day, along with other major tech stocks.
OpenAI and its primary partner, Microsoft, had already taken action against DeepSeek last year. According to a source familiar with the matter, OpenAI and Microsoft identified accounts believed to belong to DeepSeek, which had been using OpenAI’s application programming interface (API) in a manner suspected to involve distillation. These accounts were subsequently blocked for violating OpenAI’s terms. Bloomberg was the first to report on these investigations. Microsoft declined to comment on the matter, and OpenAI has not responded to further inquiries. DeepSeek has also remained silent, with no official response during the ongoing Lunar New Year holiday.
Beyond the intellectual property dispute, DeepSeek’s achievements are reshaping discussions around AI development. The company has demonstrated a method of processing vast amounts of data using significantly fewer computing resources. AI models typically learn by analysing extensive datasets, identifying patterns in text, images, and sound. DeepSeek’s approach distributes this data analysis across multiple AI models, reducing inefficiencies caused by data transfer.
The start-up’s success has major implications for the AI industry, particularly as it has chosen to release its model as open-source software. This move allows researchers and developers worldwide to build upon DeepSeek’s technology, potentially leading to the creation of uncensored AI chatbots that operate without the constraints faced by US-based firms. Some experts believe this could diminish the dominance of American AI companies, making the AI landscape more competitive.
However, concerns persist over the potential misuse of open-source AI models. Without the regulatory oversight imposed on proprietary AI systems, there is a risk that malicious actors could exploit DeepSeek’s technology to develop AI tools for disinformation or harmful activities.
Despite these challenges, DeepSeek’s breakthrough has challenged the prevailing assumption that AI supremacy requires vast financial investment. The start-up’s success demonstrates that innovation can emerge from unexpected corners, posing a significant threat to established players in the field. The impact of its advancements is already being felt, with questions being raised about whether other companies will follow suit in adopting a low-cost, open-source approach to AI development.
