Too many people still believe AI safety is a future problem that can be fixed with simple laws – but the truth is very different.