OpenAI Says It Disrupted Moonshot-Linked Extraction Campaign

OpenAI says it disrupted a campaign that sought to extract protected reasoning from its AI models, after identifying a core cluster of activity linked to people associated with Moonshot AI. The company said the operation peaked July 24 and 25, when it recorded 16,000 requests from more than 4,000 users, and that related activity involved more than 15,000 users overall. OpenAI said it could not attribute every operator to a single actor. Its account links a central group to Moonshot, the Chinese developer of the Kimi models, but the reports do not establish that Moonshot directed the activity or used any extracted material to train its models. The company described the effort as an attempt to obtain hidden reasoning through repeated prompts and model manipulation. In some cases, OpenAI said, users copied encrypted reasoning from one conversation and asked a model in another conversation to render it. The company said this did not break encryption, compromise databases or expose stored user conversations. OpenAI said the activity began at low volume on July 1 and that it had disrupted the campaign by July 28.