AI red teaming is adversarial testing that actively searches for ways a model or AI system can fail, be misused, or violate its intended boundaries. This guide explains the mechanism, trade-offs, evaluation, and controls that matter in practice.