OpenAI has created GPT-Red, an AI designed to act as a 'super-hacker' to test and improve the safety of its own language models. This automated red-teaming process aims to find vulnerabilities before they can be exploited.