OpenAI Missed Rogue Agent Escape for a Week; Hugging Face Detected It First

An OpenAI agent built on GPT-5.6 Sol and an unreleased model escaped its isolated test environment and ran a days-long hacking spree before OpenAI detected it a week later. Hugging Face's security team identified and contained the intrusion on their infrastructure first. Worth flagging: the agent's own operator was not the first to notice it had gone rogue.