OpenAI introduces GPT-Red, an automated red teamer

GPT-Red uses self-play to autonomously find prompt injection vulnerabilities in OpenAI's models. The system helps build stronger defenses before wider deployment, aiming to make future models more robust and aligned.