OpenAI introduces GPT-Red, an automated system using self-play for AI safety and robustness. This technique dynamically generates adversarial prompts to proactively identify and fix vulnerabilities, enhancing model alignment and resilience against attacks like prompt injection.
Opening Kapyn…