OpenAI claims that it has evidence that DeepSeek used its models to help train the cheaper, more efficient -V3 model.
The release of -V3 and R1 caused US tech stocks to crash on fears of the generative AI business model being upended, although many stocks began to recover the next day.
OpenAI has spent billions on its AI models, and this month launched the $500 billion Stargate effort to build even larger data centers.
DeepSeek's model was originally said to cost just $5.6m, although that was based on GPU rental figures (the company has its own compute), and only looked at the one training run.
Anthropic's CEO Dario Amodei today said that DeepSeek-V3 was not as cheap as thought, and was not something that "fundamentally changes the economics of LLMs."
Now, OpenAI and Microsoft are investigating whether DeepSeek used OpenAI’s API to integrate OpenAI’s AI models into DeepSeek’s own models. Bloomberg reports that Microsoft security researchers detected that large amounts of data were being exfiltrated through OpenAI developer accounts in late 2024, and it believes that those accounts are affiliated with DeepSeek.
OpenAI told the Financial Times that it found evidence linking DeepSeek to the use of distillation, where smaller models are built on data extracted from larger models. The company has not provided proof of its claims.
“We know PRC-based companies - and others - are constantly trying to distill the models of leading US AI companies,” OpenAI said in a statement.
“As the leading builder of AI, we engage in countermeasures to protect our IP, including a careful process for which frontier capabilities to include in released models, and believe as we go forward that it is critically important that we are working closely with the US government to best protect the most capable models from efforts by adversaries and competitors to take US technology."
President Donald Trump’s artificial intelligence czar David Sacks told Fox News that there was "substantial evidence that what DeepSeek did here is they distilled knowledge out of OpenAI models, and I don’t think OpenAI is very happy about this."
