Was DeepSeek created in part by stealing from OpenAI?

Microsoft and OpenAI launched an investigation to verify whether a group linked to the development of the Chinese AI DeepSeek improperly accessed OpenAI’s data.

According to the companies, their suspicions are based on the analyses of Microsoft’s security researchers.

It was noted that last autumn, a large amount of data had been exfiltrated via the OpenAI programming interface (API). The authors of this operation are suspected to be linked to DeepSeek.

It should be noted that the arrival of DeepSeek, revealing an unexpected level of sophistication in China, caught off guard experts, industry giants, and the American government.

The United States believed that China was lagging in AI due to restrictions on access to advanced models, limits on semiconductors, and strict regulation.

Although DeepSeek attempted to contest these claims, experts believe that if access to Microsoft’s and OpenAI’s data is confirmed, it would likely have accelerated the development of their AI.

According to the Financial Times, a source at OpenAI even claims that the company has evidence of data theft by the group.

DeepSeek AI was reportedly trained using outputs from OpenAI, a technique known as distillation.

This method allows a smaller model to learn by mirroring the responses of a more powerful model, enabling it to achieve similar performances with far fewer resources.

It should be noted that the opening of the stock markets this week was marked by a decline, prompted by the arrival of DeepSeek.

This announcement has led to an estimated loss of nearly 1 trillion dollars for American technology giants, including NVIDIA, Google, and Microsoft. Investors fear a threat to their dominance, particularly if DeepSeek has leveraged data obtained illegally to achieve such performance without the colossal investments usually required.