kapynAI / Models

OpenAI is now using AI to attack its own AI, and it's working better than humans ever did

OpenAI's internal AI, GPT-Red, now finds model vulnerabilities faster than humans. This AI-driven red teaming approach achieves an 84% success rate in identifying attacks, significantly outperforming human red teamers at 13%. The findings are crucial for enhancing the safety and robustness of future models, including GPT-5.

The Decoder·Jul 15, 2026

Opening Kapyn…