Hyderabad: Earlier this month, Chinese AI startup DeepSeek launched DeepSeek-R1, an open-source AI model, countering OpenAI's ChatGPT. The main highlight of this Chinese AI chatbot is that it offers all the premium features of ChatGPT at a cheaper rate. US President Donald Trump, who recently announced a joint venture among OpenAI and others for building AI infrastructure called DeepSeek a "wake-up call" for Silicon Valley. However, the ChatGPT maker suspects whether the newly launched Chinese AI chatbot is built upon OpenAI's data.
According to Bloomberg, OpenAI and Microsoft are investigating whether DeepSeek illicitly used OpenAI’s API to integrate OpenAI models into its own systems. As per the report, in late 2024, Microsoft security researchers detected large amounts of data being exfiltrated via OpenAI developer accounts, which is believed to be done by DeepSeek.
OpenAI told the Financial Times that it found evidence linking the Chinese artificial intelligence company DeepSeek using 'distillation'. It is a common technique through which developers extract data from larger AI models to train smaller AI models. Using this process, an AI company can efficiently train their chatbots without spending millions of dollars which OpenAI did to train its GPT-4 model. OpenAI allows developers to use its API to integrate with their applications but distilling the outputs and building a rival model violates the company's terms of service. Notably, OpenAI has not yet provided the evidence linking DeepSeek.
David Sacks, the White House AI czar spoke to Fox News on Tuesday mentioning a possibility of IP theft. “There’s 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.”
In a statement given to Bloomberg, OpenAI said, “We know PRC (China) based companies — and others — are constantly trying to distil the models of leading US AI companies.”
OpenAI said that they engage in countermeasures to protect their IP, which includes a careful process for which "frontier capabilities to include in released models" and believe that it is critically important for them to be working closely with the US government to "best protect the most capable models from efforts by adversaries and competitors to take US technology".
