OpenAI links China's Moonshot AI to model extraction

OpenAI said it identified and disrupted a coordinated campaign that attempted to extract protected reasoning from its artificial intelligence models. The company linked a core cluster of the activity to individuals associated with Moonshot AI, the Chinese startup that developed Kimi.

The activity began in early July and peaked at 16,000 requests from more than 4,000 users over two days. OpenAI ultimately identified related activity across a cluster of more than 15,000 users. The company said it had fully disrupted the operation by July 28.

OpenAI described the effort as adversarial distillation, a method where an artificial intelligence model's outputs or reasoning are extracted to train or improve another system. The company stated that operators did not breach its encryption, databases or stored user conversations. Instead, they manipulated interactions with the models to make hidden reasoning steps visible.

OpenAI noted that it remains unclear whether all participants belonged to a single actor, but it tied a central group directly to Moonshot AI. The company shared its findings with industry peers through the Frontier Model Forum and with government information channels. Moonshot AI did not immediately respond to CNBC requests for comment.