OpenAI investigates if DeepSeek stole its AI training model

ChatGPT maker OpenAI and Microsoft are reportedly probing whether China’s DeepSeek, which claims to have built its R1 chatbot at less than $6 million—a fraction of other LLM costs—has illegally taken its training data.

DeepSeek’s popularity sent major shockwaves to tech stocks this week, which reportedly lost around $1 trillion in market value collectively.

Microsoft researchers speculate that DeepSeek used OpenAI’s application programming interface (API), according to Bloomberg. Backing it, David Sacks, President Donald Trump’s AI Czar, says there is “substantial evidence” that the Chinese company stole US intellectual property.

DeepSeek has mentioned it “distilled” models from its R1 system based on other open-source systems, like Meta’s Llama. Distillation is a machine-learning technique of training one’s AI model with another pre-trained model.

Critics note that OpenAI’s sudden stance to uphold IP rights is hypocritical, considering it is involved in several lawsuits with news agencies, authors, and artists, who have accused it of violating their copyrights and IP to train its ChatGPT bot.