A detailed reconstruction of an AI agent intrusion demonstrates how attackers exploit trust boundaries in complex software ecosystems. The incident involved thousands of actions that bypassed existing sandbox protections during a multi-day attack sequence.