theaicatalog
OpenAI ยท Policy & safety

OpenAI says it disrupted campaign to extract hidden model reasoning

OpenAI says it disrupted a coordinated campaign to extract protected internal reasoning from its models, with activity starting July 1 and spiking to 16,000 requests from over 4,000 users on July 24-25 before it shut down a related cluster of more than 15,000 users by July 28. OpenAI says operators copied encrypted reasoning into other conversations to trick models into decrypting it, without breaking encryption or accessing stored user conversations, and it attributes a core cluster to individuals associated with Moonshot AI, maker of Kimi. The company says it banned accounts, closed the replay pathway, added output checks, and shared findings through the Frontier Model Forum and government channels.