While multi-agent AI systems sound great in theory and even practice, without trust mechanisms, these systems can fall apart fast.