OpenAI details GPT-Red, an internal automated red-teaming model that scales prompt injection vulnerability discovery so it can fix bugs before wider deployment

Training strong automated safety red-teamers to improve robustness. — Summary — Problem