Safety First: safety before full alignment. The deontic sufficiency hypothesis. — AI Alignment Forum