Automated Red Teaming Explained

TL;DR: AI agents don't behave the same way every time, so an attack that fails once can still work later. Automated red teaming runs each attack many times, judges what the agent actually did instead