OpenAI introduces GPT-Red, an automated red teamer for prompt injection

GPT-Red is an internal automated red teaming system that uses self-play to find prompt injection vulnerabilities at scale, turning successful exploits into training data for stronger defenses. The system aims to improve AI safety and alignment before wider deployment.