EpochLex entry

Red Teaming

red TEE-ming

Adversarial testing in which people or systems deliberately probe an AI system for harmful capabilities, weaknesses, unsafe behavior, or ways to bypass its safeguards.