OpenAI says it identifies and disrupts a coordinated campaign aimed at extracting protected “hidden reasoning” from its AI models. In a blog post, the company reports that suspicious activity starts in early July and then increases sharply in late July, leading it to shut the effort down by July 28.
OpenAI links part of the campaign to individuals associated with Moonshot AI, a Beijing-based Chinese AI developer also known for its Kimi assistant. The Next Web reports the activity begins around 1 July at low volume and spikes on 24 and 25 July, reaching roughly 16,000 requests from more than 4,000 users. CNBC similarly says OpenAI connects some activity to Moonshot AI, while The Hacker News describes the operation as a “distillation” effort to illicitly obtain protected reasoning. Quartz adds that the campaign surges and is stopped by OpenAI, framing it as an attempt to extract hidden model reasoning.
