Most alignment methods nowadays don't rely on interpretability. And neither do all LLM vendors care about alignment much - especially not in China.

Those things being untrainable at scale is why they aren't around. Alignment is an afterthought.