If frontier AI is so great, why are these frontier labs skipping security basics and making mistakes like these?

The defender problem. It's easier to attack than defend. The attacker chooses the time and the place. The defender has to win against all attacks. The attacker just needs a crack.