OpenAI Caught a Plot to Steal Its AI's Secret Reasoning

AI 📅 Oct 2, 2026 ✍️ Aamir Raza

OpenAI Caught a Plot to Steal Its AI's Secret Reasoning

16,000 requests tried to pull GPT's hidden thinking out into the open. OpenAI says it traced the operation to people linked to China's Moonshot AI. LexxWire

The Target Was GPT's Hidden Thinking

OpenAI keeps its models' step-by-step reasoning locked away from users. The campaign tried to copy that encrypted reasoning and get the model to read it back. LexxWire

16,000 Attempts in Just Two Days

The operation started quietly on July 1, then spiked July 24-25 with 16,000 extraction-style requests across 4,000 accounts. Related activity touched 15,000 users. LexxWire

Blamed on China's Moonshot AI

OpenAI attributes a core cluster to individuals associated with Moonshot AI, the Beijing startup behind the Kimi chatbot and a direct competitor. LexxWire

Shut Down, but the Trick Is Spreading

OpenAI killed the campaign on July 28, banned the accounts, and closed the replay loophole. It warns attacks like this will only get sharper. LexxWire

OpenAI Caught a Plot to Steal Its AI's Secret Reasoning