OpenAI accuses Chinese AI startup DeepSeek of using its model to train competing open-source AI

OpenAI alleges Chinese startup DeepSeek used its proprietary AI model for training a rival system.

OpenAI has accused Chinese artificial intelligence startup DeepSeek of using its proprietary models to develop an open-source competitor, raising concerns about potential intellectual property breaches in the AI industry.

The ChatGPT developer told the Financial Times that it had gathered evidence of “distillation,” a technique that allows smaller AI models to improve performance by learning from larger models. While distillation is a common practice in AI development, OpenAI claims DeepSeek may have used it to create its own competing model, which would violate OpenAI’s terms of service.

A source close to OpenAI clarified that distillation becomes problematic when developers extract model outputs and use them to build independent competitors. OpenAI’s terms explicitly prohibit users from copying its services or utilizing its output to develop rival models.

DeepSeek’s success with its R1 reasoning model has drawn attention, particularly because its efficiency contrasts with the heavy investment required by U.S. tech giants. The startup reportedly trained its V3 model using only 2,048 Nvidia H800 graphics cards and a budget of $5.6 million, a fraction of what OpenAI and Google have spent on similar models. Its ability to achieve high rankings and comparable results with significantly fewer resources has raised questions about whether it leveraged outputs from OpenAI’s GPT-4.

Last year, OpenAI and its partner Microsoft launched an investigation into accounts suspected to be affiliated with DeepSeek. They blocked access to OpenAI’s application programming interface (API) after finding indications of distillation that allegedly violated OpenAI’s terms, according to a source cited by FT.

Industry insiders note that it is common for AI labs worldwide to train models using outputs from major commercial AI platforms like ChatGPT, benefiting from human-aligned responses without incurring similar costs.

DeepSeek has not responded to the allegations, and OpenAI has declined to provide further details on its evidence. As the AI race intensifies, the dispute underscores the challenges of safeguarding proprietary technology in an industry where knowledge-sharing and competitive innovation often overlap.