“In AI development, it is more important to build robust and secure systems that raise the floor of safety and prevent disastrous outcomes than it is to achieve a perfect but potentially brittle system.”