They need to have some kind of AI safety team, as a fig leaf. But they don’t don’t want it to slow them down so they make sure it’s incompetent and ineffective.

Just a theory.