Red-teaming and evaluation

Red teaming is structured adversarial evaluation of LLM systems to find safety failures before deployment. It is not penetration testing in the traditional sense; it is a systematic exploration of the model failure modes across harm categories, attack vectors, and edge cases.