OpenAI builds GPT-Red to hack its own AI

OpenAI trained an internal AI model called GPT-Red to automatically find security flaws in GPT models via self-play reinforcement learning. GPT-Red found successful attacks in 84 percent of test scenarios versus 13 percent for human red teamers. It discovered a novel attack type called fake chain of thought. GPT-5.6 shows six times fewer failures on direct prompt injections than the best model from four months ago.