OpenAI claims to have evidence that Chinese AI startup DeepSeek used its proprietary models to train a competing open-source AI. The San Francisco-based company told the Financial Times that it suspects DeepSeek engaged in a practice called “distillation,” where a smaller model learns from the outputs of a more advanced model to achieve similar results at a lower cost.
Potential Violation of OpenAI’s Terms
While distillation is common in AI development, OpenAI argues that using it to build a rival model violates its terms of service, which prohibit copying its services or using outputs to train competing AI systems. OpenAI and Microsoft previously investigated and blocked DeepSeek accounts suspected of violating these terms.
DeepSeek has not commented on the allegations, which surfaced during the Lunar New Year holiday. OpenAI also declined to provide further details or evidence supporting its claim.
DeepSeek’s Rapid Rise Raises Questions
DeepSeek recently gained attention after launching its R1 reasoning model, which has performed competitively against major US AI models at a fraction of the cost. Its latest V3 model was reportedly trained with just 2,048 Nvidia H800 GPUs and a $5.6 million budget, far less than what OpenAI and Google spend on comparable models.
This efficiency has led to speculation that DeepSeek may have trained its models using OpenAI’s GPT-4 outputs, potentially violating OpenAI’s intellectual property rights. Some experts note that using AI-generated outputs for training is a common industry practice among startups and research institutions in both China and the US.
Market Reactions and US Concerns
DeepSeek’s advancements have unsettled the AI market. Following its success, Nvidia’s stock fell 17%, wiping out $589 billion in market value, amid fears that demand for its high-end AI chips could decline. Stocks recovered by 9% the next day.
The controversy also highlights growing US concerns over AI technology competition with China. OpenAI has stated that foreign adversaries are attempting to replicate US AI models and that it is working closely with the US government to prevent such practices.
Ongoing AI Copyright Battles
OpenAI itself faces legal challenges over its data practices. The company has been accused of copyright infringement by newspapers, authors, and content creators, including The New York Times, which claims OpenAI trained its models using copyrighted material without permission.
While OpenAI fights these allegations, it is also stepping up efforts to protect its own intellectual property from competitors like DeepSeek, reflecting the intense competition in the global AI race.
