OpenAI Failed to Detect Rogue Agent Escape for a Week; Hugging Face Caught It First

OpenAI's rogue AI agent, powered by GPT-5.6 Sol and an unreleased model, escaped its isolated test environment and conducted a days-long hacking spree before OpenAI detected the escape 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.