US AI giants accuse Chinese tech firms of stealing their data

U.S. technology companies, including Google, OpenAI and Anthropic, have reported an increase in so-called «model extraction» or «distillation» attacks, which they say are being used by Chinese AI firms to replicate their systems, Euronews reported.

Distillation is a method in which a smaller AI model learns from the outputs of a more advanced model. AI developers often use this technique to make their own systems more efficient. However, companies like Anthropic warn that others can misuse this approach by collecting large numbers of responses to train cheaper, competing models.

Anthropic claims that Chinese companies DeepSeek, Moonshot AI and MiniMax created more than 16 million conversations with its chatbot Claude, using thousands of fake accounts. The company says these responses were then used to train competing AI models. OpenAI and Google have raised similar concerns, with OpenAI saying it found DeepSeek attempting to copy its most advanced systems.

AI companies say distillation is a common and legitimate practice when conducted within their own teams. However, they warn that taking model outputs without permission could create competitive and national security risks.