The development and widespread deployment of advanced AI agents will give rise to multi-agent systems of unprecedented complexity. A new report from my colleagues at Cooperative AI Foundation and a host of leading researchers explores the novel and under-appreciated risks these systems pose.