OpenAI and Microsoft are investigating a case where DeepSeek, a Chinese artificial intelligence company, may have used data from OpenAI’s models to develop its own AI models without authorization, potentially violating service agreements.

A Bloomberg report revealed that Microsoft’s security team detected a massive data pull through OpenAI developer accounts in late 2024, which OpenAI believes may be linked to DeepSeek.

Simultaneously, OpenAI informed the Financial Times that it found evidence of the use of “distillation” techniques—a method developers employ to train AI models by extracting data from higher-capability models to reduce development costs.

Large-scale models like OpenAI’s GPT-4 require a budget of over $100 million for training, but distillation can significantly lower these costs. While developers are permitted to use OpenAI’s API to integrate AI into their own applications, using the outputs to train competing models violates service terms. However, OpenAI has not yet disclosed details of the evidence uncovered.

Read more : Meet DeepSeek, the world-changing AI from the land of the dragon, reawakening the industry

This issue is fraught with irony, as OpenAI itself initially used vast amounts of internet data without permission to develop its GPT models.

In a statement addressing the matter, OpenAI said: “We are aware that companies based in the People’s Republic of China, as well as other countries, are persistently attempting to apply distillation processes to leading U.S. AI models. As an AI leader, we have implemented safeguards to protect our intellectual property, including limiting the capabilities of models released publicly. We also view close collaboration with the U.S. government as critical to preventing powerful technology from being extracted by competitors and adversaries.”

OpenAI and its partners, including Microsoft, are now conducting further investigations to determine whether the data used by DeepSeek may violate intellectual property agreements, potentially impacting competition in the global AI industry.

Source : Bloomberg