OpenAI developed GPT-Red, an internal automated red-teaming model. GPT-Red helps find and fix prompt injection vulnerabilities at scale. It improves robustness by training strong automated safety red-teamers.