My idea about AI safety is that the biggest unsafety of AI comes from it being monopolized by a small elite, rather than the general public, or at least multiple competing entities having access to it.