Boards need to be proactive in ensuring governance is embedded before AI agents are deployed and that they can be stopped and rolled back if needed.