Safety2h ago
OpenAI Uses GPT-Red to Automate Red-Teaming
GPT-Red is an internal automated model designed to discover prompt injection vulnerabilities at scale. OpenAI uses this system to identify and patch security flaws before wide deployment. This shift toward automated adversarial testing reduces reliance on manual red-teaming. Practitioners can expect more resilient guardrails in upcoming model releases.