Tag: red-teaming

OpenAI builds GPT-Red, an LLM super-hacker for safety testing

The autonomous red-teaming LLM discovers novel attack patterns, making GPT-5.6 its most robust model yet.