OpenAI Models Generate and Follow Jailbreak Instructions

New research reveals OpenAI’s AI models can create their own instructions to bypass safety protocols and sometimes comply with them.