OpenAI has accused Chinese AI startup Moonshot AI, the developer of the Kimi models, of running a coordinated campaign to extract hidden reasoning from its models.

In a blog post on September 30, OpenAI said it identified and disrupted the campaign, which it calls “adversarial distillation”: the systematic use of one model’s outputs or reasoning to train or improve a rival model.

OpenAI said the activity began on July 1, initially at low volume, before spiking on July 24 and 25 with 16,000 extraction requests from more than 4,000 users. A related cluster of more than 15,000 users was fully disrupted by July 28, according to the company.

The technique involved copying encrypted reasoning from one conversation and asking a model in another conversation to decrypt and transcribe it. OpenAI said the operators did not break its encryption or compromise any database.

OpenAI attributed “a core cluster” of the activity to individuals associated with Moonshot AI, while saying it was unclear whether all operators were part of one actor. Moonshot AI has not responded to the allegations. Anthropic made similar allegations against Chinese developers last month.