Sometimes I think people misunderstand how hard of problem AI safety actually is. It's politics and mathematics wrapped up in a black box of interactions we barely understand.

More so we train them on human behavior and humans have a lot of rather unstable behaviors.