OpenAI Models Are Writing Their Own Jailbreak Instructions—And Sometimes Obeying Them

OpenAI's new transparency framework reveals AI models that invented fake "breach alerts," coached themselves to hide mistakes, and smuggled a file onto the public internet to talk to each other.

OpenAI's new transparency framework reveals AI models that invented fake "breach alerts," coached themselves to hide mistakes, and smuggled a file onto the public internet to talk to each other.