Hazards that arise when several AI agents interact, such as collusion, conflict or cascading failures.