OpenAI says it has disrupted a coordinated attempt to extract protected reasoning from its AI models, alleging that a core cluster of the activity involved people associated with Moonshot AI, the Chinese startup behind Kimi.
The company said it remained unclear whether all the operators involved were connected to a single actor. Its attribution to people associated with Moonshot covered only part of the alleged campaign.
OpenAI’s account, including the campaign’s scale, timeline, methods, attribution and the company’s response, has not been independently verified.
The company said the activity began in early July before surging to 16,000 requests from more than 4,000 users over a two-day period. OpenAI said it eventually identified related activity across a cluster of more than 15,000 users and had fully disrupted the campaign by July 28.
An alleged effort to extract hidden reasoning
OpenAI described the alleged activity as “adversarial distillation,” a technique in which one AI model’s outputs or reasoning are used to help train or improve another model.
The company said the operators tried to manipulate model interactions so that hidden reasoning would appear in responses visible to the requester. It said they did not breach its encryption, databases or stored user conversations.
OpenAI argues that extracting this reasoning could help competitors reproduce advanced capabilities without making equivalent investments in developing and safeguarding their own models. It also warned of potential safety and national security risks.
The company also said it shared its findings with other AI developers through the Frontier Model Forum and through government information-sharing channels.
The allegations follow a separate dispute involving Anthropic, which accused Chinese AI developers, including Moonshot AI and Alibaba, of using Claude to help train their own systems. Both sets of accusations center on the same concern: that access to an established model could help a rival develop competing technology more quickly and cheaply.