Why does every 'safety' critique about AI companies end up being completely vague like this.