Recent insights suggest that automated alignment remains a significant hurdle in AI safety research. Relying on current models to supervise even more capable systems introduces unique complexities that are often underestimated in technical roadmaps.