OpenAI Suggests DeepSeek Illicitly Used ChatGPT to Build its AI
To be clear, OpenAI has not yet come down hard saying DeepSeek definitely stole its intellectual property.

Sign up for smart news, insights, and analysis on the biggest financial stories of the day.
We regret to inform you that we have yet another story about DeepSeek, but don’t worry — this one’s pretty funny.
On Tuesday, Bloomberg reported that Microsoft and OpenAI were investigating whether DeepSeek — the AI chatbot app that took the world by storm on Monday and put a lot of egg on the face of US tech companies including OpenAI and Microsoft — had illicitly used output from ChatGPT to train its AI model. Then, on Wednesday, OpenAI told the Financial Times that it has some evidence of this being the case.
Can the Pot Sue the Kettle?
To be clear, OpenAI has not yet come down hard saying DeepSeek definitely stole its intellectual property. It told the FT it had seen some evidence of so-called distillation from accounts it suspects are linked to DeepSeek, but it declined to provide the publication with any of said evidence. Distillation is when a smaller AI model essentially gleans information from a larger model, e.g. by using text generated by the bigger model. “As a practice in research and development, it is very common and often leads to advances in the field,” Professor Brent Mittelstadt, Director of Research at the Oxford Internet Institute, told The Daily Upside.
There’s a fair amount of schadenfreude going around about OpenAI complaining that DeepSeek used text ChatGPT generated to benefit itself. After all, that’s pretty much the exact argument being leveled against OpenAI by publishers including The New York Times. However, OpenAI really can’t fall back on copyright law:
- Dr. Michael Veale, an associate law professor at University College London, told The Daily Upside that the hinted-at “distillation” isn’t really an IP issue. “OpenAI will likely find it difficult to claim copyright in the outputs from its models, so intellectual property is not going to be a very strong regime to guard against misuse of its outputs,” Veale said.
- “It is still an open question as to whether outputs of generative AI can be copyrighted or protected under IP law,” Mittelstadt said. “A consensus will likely emerge in the coming years. But for the moment, it would be incorrect to say outputs of ChatGPT or other generative models are protected under IP law by default.”
The most stalwart attack OpenAI would have is breach of contract: When you sign up for ChatGPT, you agree not to use its output for training your own AI model. In any case, Veale said, the fact the two companies are based in the US and China would complicate enforcement. Mittelstadt added DeepSeek could have been using ChatGPT legitimately: “It is possible that DeepSeek was generating outputs for a variety of other purposes, for example to benchmark their model against OpenAI’s.” So until OpenAI busts out some hard evidence, there’s no way of knowing.
Timing is Everything: President Trump’s artificial intelligence czar, David Sacks, publicly backed the theory that Chinese companies had distilled American AI tech on Fox News on Tuesday. Still, “it seems more likely that OpenAI is running this as a publicity stunt to mitigate the perception that DeepSeek has outpaced them in innovation, and instead frame themselves as still providing an essential input that the company cannot replicate without them,” Veale said.