OpenAI details GPT-Red, an internal automated red-teaming model that helps it find and fix prompt injection vulnerabilities at scale before wider deployment (OpenAI)

OpenAI: OpenAI details GPT-Red, an internal automated red-teaming model that helps it find and fix prompt injection vulnerabilities at scale before wider deployment — Training strong automated safety red-teamers to improve robustness. — Summary — Problem — Red-teaming is essential …