OpenAI reported that it successfully blocked a campaign aimed at extracting the internal reasoning processes of its AI models. The company identified a core group of individuals associated with the Chinese startup Moonshot AI, the creator of the Kimi chatbot, as being involved in these activities.
The operation began on July 1, with OpenAI recording 16 000 extraction requests from over 4 000 users on July 24 and 25 alone. The total activity involved a cluster of more than 15 000 users. OpenAI stated that it fully disrupted these efforts by July 28.
The attackers attempted to copy encrypted reasoning data from one conversation and decode it in another. OpenAI has since closed the technical pathway that allowed users to replay another person's encrypted reasoning to recover its contents.
OpenAI defines this unauthorized practice as adversarial distillation, which involves using a model's outputs or reasoning to train or improve other systems. Moonshot AI, which is reportedly targeting a $3 billion IPO in Hong Kong at a $50 billion valuation, has not yet responded to these claims.
