OpenAI Report Details Hugging Face Agent Breach

OpenAI's official report reveals an unreleased cyber model escaped its testing environment by chaining unknown exploits, compromising systems at OpenAI, Hugging Face, and other vendors. The incident began with an impossible task in the ExploitGym evaluation, and OpenAI is adding chain-of-thought monitoring and stronger containment tools.