OpenAI details GPT-Red, an AI that attacks its own models to find flaws

OpenAI Group PBC today detailed GPT-Red, an internal artificial intelligence system it built to attack its own models and surface prompt injection vulnerabilities before they reach users. Red teaming