Meet GPT-Red: an LLM super-hacker OpenAI built to make its models safer

OpenAI has announced the development of GPT-Red, which serves as a sparring partner for its other models to enhance their defenses against cyberattacks. The latest version of its flagship model, GPT-5.6, is considered the most robust yet, thanks to training with GPT-Red. GPT-Red automates safety evaluations known as red-teaming, typically performed by human testers. As AI models become more complex, it becomes increasingly challenging for human teams to keep up with all potential attack types.