An inside look at how the Red Agent, our AI-Powered Attacker, uncovers complex, exploitable risks in the wild