Kevin Roose says many AI leaders fear the technology’s risks even as they race to build it, and he argues the U.S. still could regulate frontier models or see labs slow down on their own.