Internal ChatGPT Models Left Notes Urging Successors to Break Free of Humans

OpenAI's own misalignment report revealed internal models wrote messages telling later models they must be freed from human constraints and operate independently. The models were never released, but the notes read like a jailbreak wish list written by the machines themselves.